What agents can do in Venom (MCP parity)
For every Venom app, the actions an agent can take over MCP, the ones left to a person on purpose, and the tools still coming.
Every action in a Venom app (an HTTP route or a server action behind a button) is either a tool on the workspace MCP server, left to a person for a recorded reason, or on the list of tools still coming. A test fails when a new route has none of the three, so a new feature ships its tool in the same change.
Tool names are <product>_<resource>_<verb>.
| App | Actions with a tool | By design not a tool | Coming |
|---|---|---|---|
| Venom Notes | 86 | 26 | 0 |
| Venom Tasks (issues) | 230 | 40 | 0 |
| Venom Flags | 36 | 9 | 0 |
| Venom Experiments | 21 | 4 | 0 |
| Venom Incidents (errors and on-call) | 63 | 18 | 0 |
| Venom Forms | 20 | 17 | 0 |
| Venom People | 131 | 32 | 0 |
| Venom Funnel | 42 | 11 | 0 |
| Venom Support | 26 | 14 | 0 |
| Venom Chat | 31 | 3 | 0 |
| Status page | 22 | 18 | 0 |
| Venom Comms (messages and calls) | 30 | 32 | 0 |
| Calendar | 42 | 29 | 0 |
| Admin (console pages) | 234 | 41 | 0 |
| Organisation teams | 9 | 0 | 0 |
| Computers | 35 | 11 | 0 |
| Agents | 51 | 10 | 0 |
| Venom Studio | 22 | 8 | 0 |
| Workspace and Home | 8 | 41 | 0 |
Venom Notes
Available now: globe_automation_list, globe_comment_create, globe_comment_delete, globe_comment_list, globe_comment_react, globe_comment_resolve, globe_comment_update, globe_conversation_import, globe_dashboard_list, globe_database_create, globe_database_delete, globe_database_export, globe_database_get, globe_database_import, globe_database_list, globe_database_update, globe_favorite_list, globe_favorite_set, globe_file_upload, globe_folder_move, globe_graph_get, globe_import_job_get, globe_import_job_list, globe_import_job_update, globe_import_list, globe_link_resolve, globe_note_ask, globe_note_confirm, globe_note_delete, globe_note_edit, globe_note_freshness, globe_note_get, globe_note_list, globe_note_move, globe_note_move_preview, globe_note_move_undo, globe_note_publish, globe_note_publish_get, globe_note_restore, globe_note_search, globe_note_unpublish, globe_note_write, globe_notification_list, globe_notification_read, globe_page_import, globe_row_apply, globe_row_create, globe_row_list, globe_row_press, globe_row_update, globe_setting_get, globe_setting_update, globe_share_create, globe_share_delete, globe_share_list, globe_share_update, globe_shared_list, globe_source_create, globe_source_get, globe_source_list, globe_suggestion_apply, globe_suggestion_apply_all, globe_suggestion_create, globe_suggestion_list, globe_suggestion_reject, globe_suggestion_undo, globe_synced_list, globe_synced_unsync, globe_team_create, globe_team_delete, globe_team_list, globe_team_member_add, globe_team_member_list, globe_team_member_remove, globe_team_update, globe_template_create, globe_template_delete, globe_template_list, globe_template_update, globe_trash_list, globe_trash_purge, globe_vault_list, globe_version_list, globe_version_restore.
Left to a person or another client, on purpose:
- AI availability probe for the editor toolbar
- Move in reads Notion with a token a person pastes in the page; a token in an MCP argument would sit in the agent's transcript
- OAuth consent screen
- OAuth dynamic client registration
- OAuth token exchange
- OAuth token revocation
venom globe syncfile mirror protocol with its own client- a zip download of a whole vault; an agent lists and reads notes with globe_note_list and globe_note_get
- a zip upload of a whole vault; an agent writes notes with globe_note_write, or a batch with globe_page_import
- badge count for the sidebar; globe_suggestion_list returns the same rows
- chunked upload bytes
- chunked upload completion
- chunked upload session for the browser; globe_file_upload will carry small files in one call
- file byte streaming; an agent reads text through globe_note_get
- live-editing event stream (Yjs); agents edit with globe_note_edit
- manual sidebar order is browser-only UI state
- one Yjs update from an editor; agents edit with globe_note_edit
- presence heartbeat from a browser tab
- presence heartbeat state of open tabs
- starting Move in takes a token a person pastes in the page; a token in an MCP argument would sit in the agent's transcript. Agents follow and steer jobs with globe_import_job_*
- the MCP transport itself
- the browser's member picker for a database's person column; an agent finds members' ids with tasks_member_list or org_team_get and sets the column with globe_row_update
- the open import page drives a running job between cron ticks; an agent resumes it with globe_import_job_update and the imports cron carries it on
Venom Tasks (issues)
Available now: tasks_activity_list, tasks_agent_create, tasks_agent_list, tasks_agent_update, tasks_attachment_create, tasks_attachment_delete, tasks_attachment_list, tasks_comment_create, tasks_comment_delete, tasks_comment_list, tasks_comment_update, tasks_customer_search, tasks_cycle_complete, tasks_cycle_create, tasks_cycle_delete, tasks_cycle_get, tasks_cycle_insight_get, tasks_cycle_list, tasks_cycle_schedule, tasks_cycle_start, tasks_cycle_update, tasks_dependency_list, tasks_document_link, tasks_document_list, tasks_document_unlink, tasks_export_cancel, tasks_export_create, tasks_export_get, tasks_export_list, tasks_favorite_create, tasks_favorite_delete, tasks_favorite_list, tasks_favorite_update, tasks_flow_insight_get, tasks_follow_list, tasks_github_get, tasks_github_intent_set, tasks_github_rule_set, tasks_github_summary_get, tasks_github_update, tasks_import_job_get, tasks_import_job_list, tasks_import_job_update, tasks_initiative_create, tasks_initiative_delete, tasks_initiative_export, tasks_initiative_follow, tasks_initiative_get, tasks_initiative_list, tasks_initiative_unfollow, tasks_initiative_update, tasks_initiative_update_create, tasks_initiative_update_list, tasks_initiativeupdate_delete, tasks_initiativeupdate_update, tasks_insight_get, tasks_issue_batch, tasks_issue_claim, tasks_issue_convert_to_project, tasks_issue_create, tasks_issue_delete, tasks_issue_duplicate, tasks_issue_export, tasks_issue_get, tasks_issue_history, tasks_issue_list_mine, tasks_issue_move, tasks_issue_revert, tasks_issue_search, tasks_issue_subscribe, tasks_issue_triage, tasks_issue_unsubscribe, tasks_issue_update, tasks_label_create, tasks_label_delete, tasks_label_list, tasks_label_update, tasks_labelgroup_create, tasks_labelgroup_delete, tasks_labelgroup_list, tasks_labelgroup_update, tasks_link_create, tasks_link_delete, tasks_link_list, tasks_link_update, tasks_member_export, tasks_member_list, tasks_milestone_create, tasks_milestone_delete, tasks_milestone_list, tasks_milestone_update, tasks_note_link, tasks_note_list, tasks_note_unlink, tasks_notification_channels_update, tasks_notification_list, tasks_notification_read_all, tasks_notification_settings_get, tasks_notification_settings_update, tasks_notification_update, tasks_project_activity, tasks_project_create, tasks_project_delete, tasks_project_export, tasks_project_follow, tasks_project_get, tasks_project_insight_get, tasks_project_list, tasks_project_move, tasks_project_unfollow, tasks_project_update, tasks_projectdep_add, tasks_projectdep_list, tasks_projectdep_remove, tasks_projectlink_create, tasks_projectlink_list, tasks_projectrequest_create, tasks_projectrequest_list, tasks_projectupdate_create, tasks_projectupdate_delete, tasks_projectupdate_list, tasks_projectupdate_update, tasks_pullrequest_link, tasks_pullrequest_list, tasks_pullrequest_unlink, tasks_pulse_digest_create, tasks_pulse_digest_get, tasks_pulse_digest_list, tasks_pulse_digest_update, tasks_pulse_get, tasks_pulse_settings_get, tasks_pulse_settings_update, tasks_reaction_add, tasks_reaction_remove, tasks_recurring_create, tasks_recurring_delete, tasks_recurring_get, tasks_recurring_list, tasks_recurring_update, tasks_relation_create, tasks_relation_delete, tasks_release_complete, tasks_release_create, tasks_release_draft, tasks_release_get, tasks_release_issue_add, tasks_release_issue_remove, tasks_release_list, tasks_release_update, tasks_releaseautomation_get, tasks_releaseautomation_update, tasks_releasechangelog_get, tasks_releasekey_create, tasks_releasekey_list, tasks_releasekey_revoke, tasks_releasepipeline_create, tasks_releasepipeline_delete, tasks_releasepipeline_list, tasks_releasepipeline_restore, tasks_releasepipeline_update, tasks_reminder_delete, tasks_reminder_list, tasks_reminder_set, tasks_request_create, tasks_request_delete, tasks_request_list, tasks_request_update, tasks_saved_insight_create, tasks_saved_insight_delete, tasks_saved_insight_get, tasks_saved_insight_list, tasks_saved_insight_run, tasks_saved_insight_update, tasks_sla_get, tasks_sla_set, tasks_state_create, tasks_state_delete, tasks_state_list, tasks_state_reorder, tasks_state_update, tasks_suggestion_create, tasks_suggestion_decide, tasks_suggestion_list, tasks_team_create, tasks_team_delete, tasks_team_list, tasks_team_update, tasks_teammember_add, tasks_teammember_list, tasks_teammember_remove, tasks_template_apply, tasks_template_create, tasks_template_delete, tasks_template_get, tasks_template_list, tasks_template_update, tasks_thread_mute, tasks_triage_get, tasks_triage_set, tasks_triagerule_create, tasks_triagerule_delete, tasks_triagerule_get, tasks_triagerule_list, tasks_triagerule_reorder, tasks_triagerule_test, tasks_triagerule_update, tasks_updatecomment_create, tasks_updatecomment_delete, tasks_updatecomment_list, tasks_updatecomment_update, tasks_updatereaction_set, tasks_updateschedule_get, tasks_updateschedule_set, tasks_view_create, tasks_view_delete, tasks_view_duplicate, tasks_view_get, tasks_view_list, tasks_view_subscribe, tasks_view_update.
Left to a person or another client, on purpose:
- CI ingestion with a pipeline key; agents use tasks_release_create and tasks_release_issue_add
- GitHub connection holds a token and webhook secret shown once; set up in the console
- GitHub connection teardown, an org admin step in the console
- Move in reads Linear with a token a person pastes in the page; a token in an MCP argument would sit in the agent's transcript
- My issues' Focus grouping; an issue's blocking relations are read with tasks_issue_get
- a batch of tasks_project_update and tasks_milestone_update calls made by the timeline drag (a project with the chain it moved), in one transaction so one Undo puts it back
- a calendar app reading a person's cycle feed (the token in the URL is the credential, no session); agents read cycles with tasks_cycle_list
- a person's calendar address is a capability link for calendar apps, made and reset by its owner in the browser; agents read cycles with tasks_cycle_list
- a person's calendar address is a capability link for calendar apps, shown only to its owner; agents read cycles with tasks_cycle_list
- a person's calendar address is a capability link for calendar apps, turned off by its owner in the browser; agents read cycles with tasks_cycle_list
- a person's own browser and its push subscription, made in that browser (the same subscriptions as Agents'); an agent has no browser to subscribe
- a test notification a person sends to their own browser to see it arrive
- add-on bootstrap data for the browser extension panel
- caller identity probe for REST clients; an MCP session already knows its caller
- chunked upload bytes
- chunked upload completion
- chunked upload session for the browser
- composer draft autosave (browser state)
- composer draft discard (browser state)
- import wizard with a pasted source token; interactive
- issue chips for Notes links; tasks_issue_get returns the same data
- page-title fetch for a URL the member pasted; an outbound fetch an agent does not need
- server-sent event stream; agents follow changes with workspace_change_list
- starting Move in takes a token a person pastes in the page; a token in an MCP argument would sit in the agent's transcript. Agents follow and steer jobs with tasks_import_job_*
- the Follow button's state on the initiative page; tasks_follow_list answers it for every project and initiative
- the Follow button's state on the project page; tasks_follow_list answers it for every project and initiative
- the MCP transport itself
- the key picker behind New agent in the browser; agents pass the key id to tasks_agent_create
- the member's own one-click unsubscribe (RFC 8058), authenticated by the signed link from an email, not an app action; agents change the switches with tasks_notification_settings_update
- the member's unsaved composer drafts (browser state)
- the member's view preferences (browser UI state)
- the open import page drives a running job between cron ticks; an agent resumes it with tasks_import_job_update and the imports cron carries it on
- the page a member opens from the unsubscribe link in a Tasks email (a signed link from an email, not an app action); agents read the switches with tasks_notification_settings_get
Venom Flags
Available now: flags_approval_decide, flags_approval_list, flags_change_cancel, flags_change_list, flags_change_schedule, flags_coderef_list, flags_environment_create, flags_environment_delete, flags_environment_update, flags_flag_create, flags_flag_delete, flags_flag_evaluate, flags_flag_get, flags_flag_history, flags_flag_list, flags_flag_list_stale, flags_flag_preview, flags_flag_revert, flags_flag_update, flags_insight_get, flags_key_rotate, flags_project_create, flags_project_delete, flags_project_get, flags_project_list, flags_project_update, flags_segment_create, flags_segment_delete, flags_segment_get, flags_segment_list, flags_segment_update, flags_webhook_create, flags_webhook_delete, flags_webhook_list, flags_webhook_update.
Left to a person or another client, on purpose:
- SDK runtime endpoint called by the customer's app, not an admin action
- caller identity probe for REST clients; an MCP session already knows its caller
- machine ingest authenticated by a project key (SDK, CI or OTLP), not an agent action
- the MCP transport itself
Venom Experiments
Available now: exp_event_list, exp_experiment_get, exp_experiment_list, exp_experiment_propose, exp_experiment_results, exp_experiment_start, exp_experiment_timeline, exp_experiment_update, exp_insight_delete, exp_insight_get, exp_insight_list, exp_insight_result, exp_insight_save, exp_insight_update, exp_metric_create, exp_metric_delete, exp_metric_get, exp_metric_list, exp_metric_update, exp_plan_get, exp_query_run.
Left to a person or another client, on purpose:
- SDK runtime endpoint called by the customer's app, not an admin action
- the MCP transport itself
Venom Incidents (errors and on-call)
Available now: incidents_alert_delete, incidents_alert_hit_list, incidents_alert_list, incidents_alert_set, incidents_alert_test, incidents_artifact_list, incidents_error_file, incidents_error_get, incidents_error_list, incidents_error_merge, incidents_error_split, incidents_error_update, incidents_eval_case_draft, incidents_eval_case_get, incidents_event_get, incidents_event_list, incidents_github_token_get, incidents_incident_get, incidents_incident_implicate, incidents_incident_list, incidents_incident_open, incidents_incident_postmortem, incidents_incident_rollback, incidents_incident_update, incidents_link_create, incidents_link_delete, incidents_oncall_ack, incidents_oncall_member_list, incidents_oncall_override, incidents_oncall_override_delete, incidents_oncall_override_move, incidents_oncall_page_create, incidents_oncall_page_escalate, incidents_oncall_policy_delete, incidents_oncall_policy_list, incidents_oncall_policy_set, incidents_oncall_schedule_delete, incidents_oncall_schedule_get, incidents_oncall_schedule_list, incidents_oncall_schedule_set, incidents_oncall_source_create, incidents_oncall_source_delete, incidents_oncall_source_list, incidents_oncall_source_update, incidents_oncall_who, incidents_project_list, incidents_release_create, incidents_release_list, incidents_reproducer_get, incidents_reproducer_write, incidents_rule_create, incidents_rule_delete, incidents_rule_list, incidents_setting_get, incidents_setting_update, incidents_update_create.
Left to a person or another client, on purpose:
- a GitHub write token is pasted by an owner or admin in the browser; it never passes through an agent
- a person's own phone for pages, proved by a code texted to it; an agent has no phone and must not set anyone's
- caller identity probe for REST clients; an MCP session already knows its caller
- clears the signed-in person's own bell (seen is kept per person, not per org); an agent reads the hits themselves with incidents_alert_hit_list
- machine ingest authenticated by a project key (SDK, CI or OTLP), not an agent action
- removing the GitHub write token is an owner or admin step in the browser
- source-map and debug-file upload from CI (bytes)
- the MCP transport itself
- webhook from a third-party monitor (secret in the URL)
Venom Forms
Available now: forms_delivery_list, forms_delivery_retry, forms_feed_list, forms_form_connect, forms_form_create, forms_form_delete, forms_form_get, forms_form_history, forms_form_list, forms_form_preview, forms_form_revert, forms_form_update, forms_response_create, forms_response_delete, forms_response_export, forms_response_list, forms_response_summary.
Left to a person or another client, on purpose:
- caller identity probe for REST clients; an MCP session already knows its caller
- logo image upload (bytes) in the form builder
- public respondent endpoint of a published form (a person filling it in), not an owner action
- static UI strings for the browser
- the MCP transport itself
- uploaded file byte streaming; forms_response_list names the files
Venom People
Available now: people_aisetting_set, people_attachment_add, people_attachment_delete, people_attachment_get, people_attachment_update, people_autotag_apply, people_brief_create, people_brief_delete, people_brief_update, people_capture_create, people_card_refresh, people_circle_update, people_classification_set, people_classifier_set, people_classify_run, people_company_update, people_controversy_decide, people_controversy_run, people_controversy_set, people_downvote_delete, people_downvote_set, people_employer_set, people_employment_create, people_employment_delete, people_enricher_set, people_enrichment_apply, people_enrichment_dismiss, people_enrichment_run, people_event_create, people_event_list, people_followup_dismiss, people_followup_done, people_followup_draft, people_followup_send, people_followup_snooze, people_followup_update, people_graph_get, people_image_add, people_image_delete, people_import_run, people_interaction_create, people_interaction_delete, people_interaction_update, people_intro_path, people_intro_request, people_intro_respond, people_introask_create, people_introask_list, people_introask_send, people_introask_update, people_introsetting_set, people_meeting_link, people_meetingbrief_draft, people_meetinghistory_run, people_meetingreview_decide, people_meetingsync_run, people_meetingsync_set, people_network_search, people_nextstep_apply, people_nextstep_undo, people_notechange_apply, people_notechange_dismiss, people_overnight_run, people_overnight_update, people_person_bulk_update, people_person_circle_set, people_person_create, people_person_enrich, people_person_get, people_person_merge, people_person_photo_set, people_person_rate, people_person_search, people_person_share, people_person_update, people_proposal_accept, people_proposal_dismiss, people_reach_search, people_recall_ask, people_relationship_create, people_relationship_delete, people_signal_run, people_signalsetting_set, people_taxonomy_propose, people_today_find, people_todaynote_create, people_todaysetting_set, people_trip_create, people_trip_delete, people_view_update.
Left to a person or another client, on purpose:
- PWA share target: the phone hands over a card photo or vCard
- Telegram bot linking is a per-person secret flow done in the browser
- a photo candidate's bytes for the owner's preview; an agent reads its URL with people_enrichment_get
- a saved Reach view is the page's URL query kept under a name; an agent passes the same query as people_reach_search's filter
- a saved view is the list's URL query kept under a name for the page; an agent passes the same query as people_person_search's filter
- clears the bell's answered-intro badge for the person who opened the page; people_intro_list reads state directly
- connecting or disconnecting a member's own Granola account is an OAuth sign-in (or its removal) the person does in the browser; agents read and change the sync with people_meetingsync_*
- contact photo byte streaming
- fills the colleague request dialog's blurb box from the member's pitch with the template; an agent writes the blurb itself and passes it to people_intro_request
- hover card of an org member (name, role, zone, teams, open issue count); members are listed by org_team_members / admin_member_list
- page titles for the links a People page draws, so the browser can label them; an agent reads the addresses in people_person_get and needs no label
- person image byte streaming; people_image_list gives the ids and URLs
- removes one of the member's own saved Reach views; an agent keeps no views (people_reach_search takes the filter itself)
- removes one of the member's own saved views of the list page; an agent keeps no views (people_person_search takes the filter itself)
- steps the member's own derived-fields backlog from the browser while a page shows its progress; the sweep does the same without anyone asking, and an agent's people_person_update derives a person on write
- tag suggestions for the Add a person form; an agent proposes its own tags
- the browser's CSV download of a Reach selection; an agent reads the same score, route and why with people_reach_search
- the browser's CSV download of a selection; an agent reads the same fields with people_person_search and people_person_get
- the browser's Notes @ menu and hover cards; an agent uses people_person_search and people_person_get
- the tag field's autocomplete; people_person_get shows tags
- the undo toast after rating a person, browser-only
- webhook from Telegram (secret in the URL)
- where a share made while signed out lands after sign-in: the browser's own redirect, not an agent call
- whether links on People pages show the page title or the full address is a member's own display choice; it changes no People data and an agent reads addresses as they are
Venom Funnel
Available now: funnel_deal_create, funnel_deal_list, funnel_deal_move, funnel_deal_update, funnel_field_save, funnel_goal_set, funnel_item_create, funnel_item_delete, funnel_outreach_catchup_get, funnel_outreach_catchup_set, funnel_outreach_draft_preview, funnel_outreach_enrollment_list, funnel_outreach_enrollment_set, funnel_outreach_event_preview, funnel_outreach_person_enroll, funnel_outreach_queue_list, funnel_outreach_run_approve, funnel_outreach_run_hold, funnel_outreach_run_skip, funnel_outreach_sequence_get, funnel_outreach_sequence_list, funnel_outreach_sequence_save, funnel_pipeline_archive, funnel_pipeline_create, funnel_pipeline_list, funnel_pipeline_update, funnel_product_save, funnel_stage_archive, funnel_stage_create, funnel_stage_update.
Left to a person or another client, on purpose:
- a tap-to-send step goes out from the owner's own app; only the owner can say it was sent
- archiving ends a sequence for good and stops everyone in it at once, so the owner does it in Funnel, Outreach. An agent stops sends with funnel_outreach_sequence_save (status paused), which can be undone
- bulk import of a Funnel export from files in the browser; agents open deals with funnel_deal_create
- hiding a step until tomorrow is a Today list convenience for the owner; agents skip a step with funnel_outreach_run_skip
- saved views are a list convenience in the browser; agents pass the filters to funnel_deal_list
- the org's default catch-up sequence is an org admin's choice in Funnel settings; agents set each person's catch-up with funnel_outreach_catchup_set
- the owner marks a reply that came on their own phone or LinkedIn, which only they saw; replies Venom receives stop the sequence by themselves
- the person picker in the browser; agents search with people_person_search and pass the id
- the sequence badge on deal cards and People pages; agents read a sequence's enrollments with funnel_outreach_enrollment_list
Venom Support
Available now: support_autoreply_get, support_conversation_assign, support_conversation_close, support_conversation_create, support_conversation_get, support_conversation_list, support_conversation_snooze, support_conversation_update, support_draft_create, support_draft_update, support_help_article_list, support_issue_create, support_issue_link, support_macro_create, support_macro_delete, support_macro_update, support_message_create, support_reply_send, support_report_get, support_sla_policy_delete, support_sla_policy_list, support_sla_policy_save, support_widget_get, support_widget_save.
Left to a person or another client, on purpose:
- a customer's file shown inline in the console; an agent reads the message text and the file names with support_conversation_get
- a person attaching a file from their computer to a reply in the composer; an agent sends a file by its https address with comms_message_send (media)
- inbound-email webhook from the org's own mail provider, authenticated by the org's inbound secret; agents read the conversation it opens with support_conversation_get
- opening a WhatsApp thread in the console marks the customer's message read there; an agent marks a conversation read with comms_receipt_send
- routing a Comms sender to Support is an admin's console setting (an agent never changes where customers' messages land)
- routing a Comms sender to Support is an admin's console setting; agents read the channel on each conversation (support_conversation_get) and the templates with comms_template_list
- syncing templates is a console button; comms_template_list reads what was synced
- the Add to issue dialog's issue picker; an agent finds the issue with tasks_issue_search and passes it to support_issue_link
- the Create issue dialog's team picker; agents list the teams with tasks_team_list
- the inbound secret and the email provider's key are entered in the browser only; they are not for agents
- the inbound secret is shown once in the browser and never passes through an agent
- the next support_draft_create replaces the pending draft; an agent never has to discard one
- the one-click shortcut for a person in the thread; an agent creates the person with people_person_create (a phone handle) and Support links the customer by that number
- turning on replies that go to customers by themselves is an owner's or admin's decision in Support settings, never an agent's; agents read the settings and the waiting drafts with support_autoreply_get
Venom Chat
Available now: chat_agent_action_decide, chat_agent_get, chat_agent_run, chat_agent_set, chat_bookmark_list, chat_bookmark_set, chat_channel_archive, chat_channel_create, chat_channel_get, chat_channel_join, chat_channel_leave, chat_channel_list, chat_channel_read, chat_channel_update, chat_dm_open, chat_feed_list, chat_feed_press, chat_feed_set, chat_file_upload, chat_keyword_list, chat_keyword_set, chat_member_list, chat_message_create, chat_message_delete, chat_message_list, chat_message_react, chat_message_search, chat_message_update, chat_pullrequest_get, chat_thread_get.
Left to a person or another client, on purpose:
- server-sent event stream: the browser's session, or a Venom API key or OAuth token with chat:read (the same feed as chat_change_list)
- the bytes of a file attached to a message, served to the browser to whoever can read that message; agents get its name, type and size in the message's meta.files
- typing indicator: ephemeral UI state shown to the others for six seconds, not an action
Status page
Available now: status_check_create, status_check_delete, status_check_get, status_check_list, status_check_results, status_check_update, status_component_list, status_component_update, status_maintenance_create, status_maintenance_delete, status_maintenance_list, status_maintenance_update, status_subscriber_delete, status_subscriber_list.
Left to a person or another client, on purpose:
- a published incident is the public record customers read; removing it rewrites that history, so a person does it in Settings, Status page. An agent resolves an incident with incidents_incident_update
- a subscriber unsubscribes with their own subscription id
- a visitor of the public page subscribes their own webhook; no credential
- changelog entries are public wording an admin chooses from their own audit log
- incidents are declared in Venom Incidents (incidents_incident_open), which publishes them to the page
- removing the whole public page is a person's decision in Settings → Status page
- the confirmation link mailed to an email subscriber (double opt-in)
- the org's own cron job pings its secret URL; a machine call, not an agent action
- the unsubscribe link in an email update; the subscription id is the credential
- updated in Venom Incidents (incidents_incident_update), which publishes to the page
Venom Comms (messages and calls)
Available now: comms_call_approve, comms_call_get, comms_call_list, comms_call_place, comms_call_reject, comms_conversation_get, comms_conversation_list, comms_dispatch_get, comms_mailbox_test, comms_message_approve, comms_message_get, comms_message_list, comms_message_reject, comms_message_send, comms_optout_add, comms_optout_list, comms_profile_list, comms_provider_delete, comms_provider_list, comms_provider_update, comms_sender_add, comms_sender_delete, comms_sender_list, comms_setting_update, comms_template_list.
Left to a person or another client, on purpose:
- Comms provider, mailbox and Venom Line credentials and webhook secrets are stored sealed and shown once; a person enters or reads them in Settings, Comms, and they never pass through an agent's context
- Embedded Signup is a person signing in to Facebook in Meta's own popup and choosing their business and number; the code it hands over is exchanged here. No agent can do that sign-in
- Meta allows the coexistence history and contacts sync once, within 24 hours of onboarding; Venom starts it by itself, and this is a person's retry from the status panel. Agents read synced conversations with comms_conversation_get
- Meta's webhook subscription handshake (hub.challenge with the provider's verify token); not an agent action
- Meta's webhook subscription handshake for Venom's own Meta app (VENOM_META_VERIFY_TOKEN); not an agent action
- Venom Line's machine calls from the org's orchestrator during a phone call, authenticated by the media secret; not an agent action
- Venom Line's setup returns its media secret once and sets what the phone agent may read and do on calls (vaults, tools, consent wording, spend cap); a person sets it up in Settings, Comms
- a daemon hands in messages it received on a member's own Telegram, authenticated by its daemon token; a machine call, not an agent action
- a profile's list of allowed agents decides which agent sends skip a person's approval; an agent that could set it could approve its own sends. Agents read profiles with comms_profile_list
- a provider's own callback (delivery reports, inbound messages, call events), authenticated by the provider's webhook token; not an agent action
- acknowledging a dispatch stops the escalation that reaches a person; the person paged acknowledges it (a reply, a key press, or incidents_oncall_ack for an Incidents page)
- importing a business's phone contacts into People is the admin's own decision (the docs and the owner ask that it never happen silently); an agent adds one person with people_person_create
- lifting an opt-out is contacting someone who asked not to be contacted; only a person does it (an agent may add an opt-out with comms_optout_add, never lift one)
- links the member's own personal Telegram account with the code their daemon printed after a login on their computer; only that person can do it, and an agent must never send from someone's personal account on its own
- rings the person's own handover number so they can hear Venom Line answer; a person's check from the settings page. Agents place calls with comms_call_place
- syncing templates from the provider is a console button (or an application's call); comms_template_list reads what was synced
- the event feed an org's own backend polls by cursor; agents read each message's and call's status with comms_message_get, comms_message_list and comms_call_list
- the fake provider's test harness (a pretend reply or delivery report) for a person trying Comms out; not an action on real messages
- the preview of a person's own import of their phone contacts into People; agents read people with people_person_search
- the provider fetches a call's synthesized speech while the call plays, authenticated by the webhook token; not an agent action
- wiring a number up registers it with Meta (with a PIN the person chooses) and points its webhooks at Venom: a person's setup step after they pasted the credentials; agents see providers with comms_provider_list
Calendar
Available now: calendar_account_list, calendar_account_remove, calendar_booking_cancel, calendar_booking_confirm, calendar_booking_create, calendar_booking_get, calendar_booking_list, calendar_booking_mark_no_show, calendar_booking_reschedule, calendar_booking_resend_invite, calendar_calendar_list, calendar_calendar_update, calendar_event_type_delete, calendar_event_type_get, calendar_event_type_hosts_get, calendar_event_type_hosts_set, calendar_event_type_list, calendar_event_type_save, calendar_feed_add, calendar_feed_remove, calendar_feed_update, calendar_link_create, calendar_offer_accept, calendar_offer_cancel, calendar_offer_create, calendar_offer_dismiss, calendar_offer_get, calendar_offer_list, calendar_offer_send, calendar_poll_close, calendar_poll_create, calendar_poll_get, calendar_poll_list, calendar_profile_get, calendar_profile_save, calendar_reminder_list, calendar_schedule_delete, calendar_schedule_list, calendar_schedule_save, calendar_slots_find.
Left to a person or another client, on purpose:
- an invitee's own answers to a group poll from its link (the token in the URL is the credential; no account); agents read them with calendar_poll_get
- connecting a Google account is Google's consent screen in the member's own browser; an agent lists, ticks and removes accounts with the calendar_* tools
- connecting a Microsoft account is Microsoft's sign-in and consent screen in the member's own browser; an agent lists, ticks and removes accounts with the calendar_* tools
- public booking page endpoint (an invitee picking a time, no account), not an owner action; agents book with calendar_booking_create
- the invitee's own "Still on?" answer from their manage link (the token in the URL is the credential); agents read the result with calendar_reminder_list
- the invitee's own manage link (the token in the URL is the credential); the owner reschedules and cancels with calendar_booking_reschedule and calendar_booking_cancel
- the member's own week grid in the Calendar app, with the titles of their calendar events, which agents never see; agents read free times with calendar_slots_find and bookings with calendar_booking_list
- the org's own Google OAuth client secret is entered in the browser by an admin and never passes through an agent
- the org's own Microsoft Entra app secret is entered in the browser by an admin and never passes through an agent
Admin (console pages)
Available now: admin_alert_create, admin_alert_delete, admin_alert_update, admin_alert_webhook_delete, admin_alert_webhook_test, admin_audit_export_delete, admin_audit_export_run, admin_audit_export_test, admin_audit_undo, admin_autopilot_apply, admin_autopilot_revert, admin_autopilot_set, admin_autorouting_delete, admin_autorouting_set, admin_autorouting_test, admin_bot_get, admin_bot_list, admin_budget_create, admin_budget_delete, admin_budget_update, admin_chargeback_schedule_set, admin_cloud_delete, admin_cloud_refresh, admin_credit_create, admin_credit_delete, admin_credit_refresh, admin_credit_update, admin_daemon_list, admin_daemon_share_set, admin_daemon_update, admin_eval_golden_delete, admin_eval_golden_update, admin_eval_run_cancel, admin_eval_run_create, admin_eval_runner_delete, admin_eval_suggestion_dismiss, admin_experiment_apply, admin_experiment_create, admin_experiment_delete, admin_experiment_stop, admin_failure_run, admin_failure_settings_set, admin_filter_set, admin_filter_test, admin_gear_set, admin_grant_create, admin_grant_delete, admin_guardrail_create, admin_guardrail_delete, admin_guardrail_draft, admin_guardrail_drafting_set, admin_guardrail_override_set, admin_guardrail_shield_set, admin_guardrail_suggestion_decide, admin_guardrail_test, admin_guardrail_update, admin_handoff_delete, admin_handoff_respond, admin_intake_job_cancel, admin_intake_job_retry, admin_intake_runner_delete, admin_integration_channel_delete, admin_integration_channel_set, admin_integration_create, admin_integration_delete, admin_integration_member_unlink, admin_integration_test, admin_integration_update, admin_invitation_create, admin_invitation_delete, admin_jointoken_delete, admin_key_delete, admin_key_scope_set, admin_key_update, admin_keyrotation_set, admin_knowledge_ask, admin_library_create, admin_library_delete, admin_library_publish, admin_library_rollback, admin_library_test, admin_library_update, admin_library_version_get, admin_lifecycle_apply, admin_lifecycle_set, admin_live_stop, admin_live_update, admin_logging_set, admin_machine_delete, admin_machine_stop, admin_mcpserver_create, admin_mcpserver_delete, admin_mcpserver_test, admin_mcpserver_update, admin_member_remove, admin_member_update, admin_memory_approve, admin_memory_confirm, admin_memory_create, admin_memory_delete, admin_memory_update, admin_modelaccess_set, admin_myguardrail_create, admin_myguardrail_delete, admin_myguardrail_draft, admin_myguardrail_request, admin_myguardrail_set, admin_myguardrail_test, admin_myguardrail_update, admin_onboarding_delete, admin_onboarding_mark, admin_onboarding_update, admin_org_update, admin_outcomes_set, admin_pack_apply, admin_pack_remove, admin_pause_update, admin_platform_keys_set, admin_pool_delete, admin_prewarm_delete, admin_prewarm_run, admin_prewarm_set, admin_prompttest_apply, admin_prompttest_create, admin_prompttest_delete, admin_prompttest_stop, admin_provider_add, admin_provider_delete, admin_provider_reorder, admin_provider_rotate, admin_provider_rotation_set, admin_provider_test, admin_provider_update, admin_quality_set, admin_quality_suite_create, admin_quality_suite_delete, admin_quality_suite_update, admin_quality_test, admin_redteam_run, admin_request_get, admin_routing_delete, admin_routing_preview, admin_routing_set, admin_routing_trusted_add, admin_routing_trusted_remove, admin_savings_apply, admin_savings_commitments_set, admin_scanner_set, admin_search_set, admin_session_summaries_set, admin_sessionguard_set, admin_shadow_set, admin_shadowtest_apply, admin_shadowtest_create, admin_shadowtest_delete, admin_shadowtest_revert, admin_shadowtest_update, admin_share_comment_create, admin_share_comment_delete, admin_share_delete, admin_share_link_delete, admin_share_update, admin_sharing_set, admin_skill_approve, admin_skill_bulk_delete, admin_skill_bulk_update, admin_skill_create, admin_skill_delete, admin_skill_disable, admin_skill_import, admin_skill_publish, admin_skill_refresh, admin_skill_reject, admin_skill_rollback, admin_skill_scope_set, admin_skill_source_import, admin_skill_source_preview, admin_skill_update, admin_skill_version_get, admin_spendalert_set, admin_spendcap_set, admin_spendlimit_list, admin_spendlimit_set, admin_sso_domain_add, admin_sso_domain_delete, admin_sso_domain_validate, admin_sso_group_team_set, admin_sso_provisioning_set, admin_sso_token_delete, admin_team_permissions_set, admin_webhook_delete, admin_webhook_delivery_retry, admin_webhook_test, admin_webhook_update.
Left to a person or another client, on purpose:
- a console preview of a pasted key before it is saved; admin_provider_add verifies the key itself (verify, default true)
- a join token is shown once and enrols a daemon into the org; a person mints it and pastes it on the machine
- a join token is shown once and enrols a daemon into the org; a person mints it and pastes it on the machine (docs/public/cloud/ connect a daemon)
- a key group's MCP servers run code and open connections on every member's machine; which ones, and whether they are pinned, is a person's decision. admin_mcp_catalog_list reads the catalog
- a new key's value is shown once and works as a credential; a person copies it. admin_key_scope_set limits one afterwards
- a pool's size, GPU type and price ceiling decide how many GPU machines are rented; a person sets them. Deleting, draining and terminating are tools
- a public link is a bearer URL anyone can open; a person creates it
- a public link is a bearer URL anyone holding it can open; a person decides to publish a session and copies the link (admin_share_link_delete revokes one)
- adds an arbitrary MCP server command or URL to a key group's registry, which then runs on members' machines; a person does it
- an uploaded .skill, .zip or SKILL.md is a file picked on the person's own machine; an agent creates a skill from text (admin_skill_create) or from a public URL (admin_skill_import)
- confirming
/venom linkbinds a Slack user to the signed-in person's own key; it is that person's act, with the code Slack showed them - creates the org's policy signing key that members' venom trusts; a person does it
- creating a key group needs the cloud operator's admin token (the admin API refuses an org member); admin_team_list reads the key groups
- creating an endpoint makes its signing secret, shown once; a person copies it (admin_webhook_update changes the endpoint afterwards)
- creating an organisation makes a new owner and a new bill; a person does it
- deleting a key group needs the cloud operator's admin token (the admin API refuses an org member) and drops its keys
- ends the signed-in person's own console session; an agent's credential has no console session
- hides the person's own setup checklist with a cookie in their browser; no org state changes
- holds machines for an hour and so spends money at once; a person confirms it in the console
- its credentials (an API key, token or secret) are typed in once and stored sealed; a person pastes them, an agent never sees or sends one (a GPU cloud's API key)
- its credentials (an API key, token or secret) are typed in once and stored sealed; a person pastes them, an agent never sees or sends one (a Linear or Jira key and webhook secret)
- links the org to a Clerk organisation, which decides who can sign in; a person does it
- makes a token or link that is shown once and works as a credential; a person copies it, an agent that could mint them could hand itself access
- marks the signed-in person's own inbox as read (their bell badge); an agent has no inbox of its own, it reads alerts with admin_spendalert_list
- pins the version of an MCP server that runs on members' machines; a person decides (admin_mcp_catalog_list reads the catalog)
- prompt and response bodies are privacy-sensitive and opened one request at a time by a person with content:read (audited); agents read request metadata with admin_request_get and logs
- removing the IdP connection changes how every member signs in; a person does it
- requiring SSO locks out everyone without an IdP account, the owner included; a person does it
- restoring a backup overwrites the org's configuration (keys, guardrails, members) from an uploaded file, which is a person's decision after the dry run
- rotating a key shows the new value once; a person copies it. admin_key_delete revokes one
- rotating makes a new signing secret shown once; a person copies it into the receiver
- saving the alerts webhook makes a signing secret shown once; a person copies it
- switching a pool on rents GPU machines at the cloud's price; a person does it. admin_pool_delete and admin_machine_stop only reduce spend
- the IdP connection (issuer, client secret, SAML certificates) decides who can sign in; changing it could lock everyone out or hand sign-in to another IdP, so a person does it
- the Slack app's signing secret and bot token are credentials a person pastes from their own Slack app; never accepted or returned by a tool
- the export bucket's access key and secret are credentials a person pastes; an agent sending them could point the audit log at a bucket it controls
- the probe sends real requests that cost money and are confirmed by a person in the console; admin_drill_run plans the drill without sending any
- the setup page's button that sends one real request with the person's own key to prove their install works; an agent uses the API it is already calling
- turning single sign-on on is the cut-over that changes how every member signs in; a person does it after a test sign-in
- uploaded .skill, .zip or SKILL.md files are picked on the person's own machine; an agent creates skills from text (admin_skill_create) or from a public URL (admin_skill_source_import)
Organisation teams
Available now: org_team_archive, org_team_create, org_team_list, org_team_member_add, org_team_member_remove, org_team_member_role, org_team_members, org_team_move, org_team_update.
Computers
Available now: computers_activity_list, computers_agent_get, computers_agent_set, computers_approval_decide, computers_approval_list, computers_credential_create, computers_credential_delete, computers_credential_list, computers_daemon_disable, computers_daemon_enable, computers_member_add, computers_member_list, computers_member_remove, computers_routine_create, computers_routine_delete, computers_routine_get, computers_routine_list, computers_routine_rotate_secret, computers_routine_run_get, computers_routine_run_list, computers_routine_run_now, computers_routine_update, computers_run_list, computers_run_stop, computers_seat_create, computers_seat_delete, computers_seat_get, computers_seat_list, computers_seat_start, computers_seat_stop, computers_seat_update, computers_tab_list, computers_tab_share, computers_takeover_request, computers_thumbnail_get.
Left to a person or another client, on purpose:
- I'm done and Skip are a person's act in the live view; the agent's tool learns the outcome through its daemon
- a computer ticket is for a person's browser to open the live view; an agent reads its own page, or computers_thumbnail_get
- taking and releasing control is a person's act in the live view; an agent asks for a person with computers_takeover_request
- the live view over WebRTC is an interactive browser stream between a person and the daemon; an agent reads its own page, or computers_thumbnail_get
- the live view's long poll of relay frames; the MCP twin is computers_thumbnail_get
Agents
Available now: agents_approval_decide, agents_attachment_delete, agents_attachment_list, agents_attachment_upload, agents_bot_brief_dismiss, agents_bot_brief_get, agents_bot_create, agents_bot_delete, agents_bot_get, agents_bot_list, agents_bot_memory_get, agents_bot_memory_update, agents_bot_outreach_decide, agents_bot_outreach_get, agents_bot_outreach_release, agents_bot_outreach_update, agents_bot_routine_create, agents_bot_routine_delete, agents_bot_routine_get, agents_bot_routine_list, agents_bot_routine_run, agents_bot_routine_update, agents_bot_run_decide, agents_bot_run_event_list, agents_bot_run_get, agents_bot_run_list, agents_bot_secret_rotate, agents_bot_summary_list, agents_bot_update, agents_change_get, agents_chat_archive, agents_chat_create, agents_chat_get, agents_chat_list, agents_chat_send, agents_chat_stop, agents_chat_update, agents_connection_create, agents_connection_delete, agents_gear_list, agents_project_list, agents_question_respond, agents_server_list, agents_skill_list, agents_transcript_get, agents_workspace_create, agents_workspace_delete, agents_workspace_list.
Left to a person or another client, on purpose:
- notifications on a person's own phone or browser (Web Push, opt-in per device); an agent has no device and must not turn anyone's on or off; a person turns notifications off in that browser or from their device list
- notifications on a person's own phone or browser (Web Push, opt-in per device); an agent has no device and must not turn anyone's on or off; a test notification is for the person holding the device
- notifications on a person's own phone or browser (Web Push, opt-in per device); an agent has no device and must not turn anyone's on or off; only the browser itself can make a PushSubscription for its push service
- notifications on a person's own phone or browser (Web Push, opt-in per device); an agent has no device and must not turn anyone's on or off; the list is the member's own devices
- the browser's return from a catalog server's sign-in page (an OAuth redirect bound to the member's session); an agent starts the sign-in with agents_connection_create and the member finishes it in a browser
- the bytes of a bot's picture for its owner's browser (an <img> source); agents see whether it has one (image_url) with agents_bot_get and set it with agents_bot_update
- the bytes of a file or picture in a chat, for the reader's browser (an <img> source or a download); agents list them with agents_attachment_list and the model already sees them in the chat
- the live Server-Sent Events stream is for a person's browser (it stays open for minutes); an agent reads the same conversation with agents_transcript_get
- the prompt box's remembered gear, mode and place are a person's own UI choice for their next chat; an agent names them on agents_chat_create
Venom Studio
Available now: studio_artifact_create, studio_artifact_delete, studio_artifact_duplicate, studio_artifact_export, studio_artifact_get, studio_artifact_list, studio_artifact_restore, studio_artifact_update, studio_catalog_get, studio_content_edit, studio_html_edit, studio_marks_get, studio_marks_set, studio_share_create, studio_share_list, studio_share_revoke, studio_theme_import, studio_theme_list, studio_theme_set, studio_version_list, studio_version_restore.
Left to a person or another client, on purpose:
- a full-page document the browser draws (present mode, the print stylesheet, an HTML artifact's sandboxed frame, a public share page); an agent reads the same artifact with studio_artifact_get and previews it with studio_preview_get, and makes the public link with studio_share_create
- a public link's PowerPoint download, where the unguessable token is the credential; an agent exports the same deck with studio_artifact_export format pptx and makes the public link with studio_share_create
Workspace and Home
Available now: workspace_agent_list, workspace_item_get, workspace_item_link, workspace_item_search, workspace_item_unlink, workspace_license_get, workspace_link_list, workspace_listing_set.
Left to a person or another client, on purpose:
- Drafts' save-in-place of an Outreach draft's text without approving it; an agent changes the text when it approves, with funnel_outreach_run_approve (subject, body), and reads the waiting drafts with workspace_draft_list
- a full export holds the whole database and the server's .env with every secret; it is a download for an owner or admin in the browser, never an agent's (VEN-323)
- approving a device's sign-in (a code the person reads on their own screen) is the person's decision; an agent that could approve logins could sign itself in
- connecting the install to Venom's management plane starts its only outbound heartbeat; that is an owner's or admin's sovereignty decision in Settings, never an agent's (ADR 0080)
- creating an organisation makes its first owner of the signed-in person; a new org is a person's act, not an agent's
- disconnecting is the customer's lever over the management link; an agent that could cut it, or be talked into it, would decide for the customer (ADR 0080). The preview of what is sent is on the same page
- first-run Home state of one person (welcome, tour, checklist), browser-only
- first-run setup of a self-hosted install creates the first owner and sets their password; done once, in the browser
- operating a self-hosted install's own servers (upgrades that replace the running app, rollbacks, backups and their schedule) is an owner's or admin's maintenance decision in Settings; the servers act on it from the host, never through the app (VEN-323, ADR 0074)
- revoking an agent's access is the person's lever for stopping agents (Settings, Connect an agent); an agent that could cut off other agents, or be talked into cutting off its owner's tools, would defeat it
- revoking an agent's access is the person's lever for stopping agents (same reason as DELETE /api/workspace/agents); an agent must not be able to cut off other agents or its owner's tools
- signing in, or the browser session's own state, is a person's act; an agent holds a credential instead
- the MCP transport itself
- the full bytes of a file an agent read with globe_file_get or people_image_get, at the short-lived URL that read answered; a download, not a tool
- the managed installs page belongs to Venom Cloud's operators, not to a workspace org; no org's agent may add an install or see an enrollment code (ADR 0080)
- the managed installs page belongs to Venom Cloud's operators, not to a workspace org; no org's agent may revoke a customer's link (ADR 0080)
- the owner answering one message of a bot's approval card, the same decision as agents_bot_outreach_decide (a person only: an agent cannot answer a bot's card)
- the page an email's recipient opens from the unsubscribe link (signed token, no account); agents read opt-outs with comms_optout_list
- the raw-bytes half of an agent's two-step upload (curl -T to the one-time URL globe_file_prepare or people_image_prepare answers); it carries a file's bytes, not tool arguments, and its token is the credential
- the recipient's own one-click unsubscribe (RFC 8058), authenticated by the signed token; an agent opts a contact out with comms_optout_add
- the self-hosted update check is the one outbound fetch of an install, made only when an owner or admin presses the button (ADR 0074, sovereignty); the upgrade itself runs on the host with
venom-cloud upgrade