What your assistant can do
The tools a connected assistant or agent gets over MCP, family by family — and what is deliberately absent.
A connected assistant works through a fixed set of tools — sixty-six of them, the same rules for every client: a token limited to reading sees only the reading tools — plus attachable Guidance and documents, and the role prompts. Every tool obeys the same rules as the app: your permission, the project’s workflow, and the draft policy. This page lists the tools by family, in plain words, and then what is missing on purpose.
Finding its way
list_projects · get_identity · get_project_structure · get_workflow · list_custom_fields · list_types · list_members · list_agents · get_guidance · get_context
The assistant learns which projects it can reach and as whom, how a project is organised, which statuses and transitions exist, which fields and task types are defined, who is on the project, and the project’s Guidance — all before it touches anything. get_context gathers what it needs about one task — its parents, children, linked pages, dependencies — in one call.
Tasks
list_tasks · get_task · get_task_counts · create_task · update_task · edit_task · add_dependency · remove_dependency · list_dependencies · set_task_sprint · link_task_release · unlink_task_release
File tasks, read and search them, change their status (only along the transitions the workflow allows), edit their description, link dependencies, place a task in a sprint or take it out, and attach it to a release or detach it.
Documents
list_documents · get_document · create_document · update_document · edit_document · move_document · preview_markdown · link_document_task · unlink_document_task · upload_asset · get_asset
Read pages as markdown and write them back as markdown, create and move pages in the tree, check how a piece of markdown will be stored, link a page to a task, attach images and files, and read images back.
Comments
add_comment · list_comments · resolve_comment
Comment on tasks and pages, read the discussion, and resolve a thread.
Reviews and revisions
start_review · review_scope · add_review_comment · set_finding_disposition · complete_review · list_review_passes · submit_revision · get_revision · get_revision_diff · get_revision_history · approve_revision · reject_revision
Run a review pass over a page — findings as comments, a closing summary — propose a revision as a draft, read a page’s history and the diff between any two revisions, and, for a person with Member access or higher acting through their own assistant, approve or reject a pending draft. A project agent never approves.
Tests
list_test_cases · create_test_case · update_test_case · create_test_run · record_test_result · get_test_run · list_test_runs
Write and update test cases, start a run, and record what passed and failed.
Sprints and releases
list_sprints · get_sprint_report · list_releases · get_changelog
Read the sprints and their reports, the releases, and the changelog of a release — the answers to “how did the last sprint go?” and “what is in 2.1?” come from here.
Search and watching
search · watch_entity · unwatch_entity · list_watched_changes
Search the project in any of its languages, follow a task or a page or stop following it, and read the status changes, comments, revisions and reviews on the items it follows.
Translations
list_translations · submit_translation · list_translation_gaps
Read a page’s translations, propose one as a draft, and work the list of what is missing or outdated per language.
What is absent on purpose
There is no tool to delete a task, a page, a comment or a test case; the remove and unlink tools above take a link away and leave both ends standing. There is no tool to change a project’s settings, workflow, members, roles or agents, to publish a skill, to bind a Library, or to export. Those are a person’s acts in the app. An assistant that is told to do one of them has no MCP tool to do it with.
Two more rules hold on writes: a body edit — a task’s description, a page, a test case’s steps — carries the revision it was based on and is refused if that moved underneath, and an edit to existing content may wait as a draft — see Review what your AI proposes. Calls are rate-limited per token and per project; a refused call says when to try again.