AI assistants (MCP)

Tools reference

Every tool an assistant can call, and the permission each one needs.

An assistant reads this list from the server, so it always matches what your token allows.

Projects — projects:read

Tool What it does
list_projects The projects this token can reach. Start here: every other tool takes a project id.
get_project One project, with how many cases, flows and milestones it holds.

Test cases — cases:read, cases:write

Tool Needs What it does
search_cases read Find cases by words, tag, status or kind.
get_case read One case in full, by id or reference (TC-0004).
create_case write Create a manual case. It starts as Draft. A duplicate name is refused.
update_case write Change a manual case. Passing steps replaces all of them.
list_flows / save_flow read / write Flows and their cases.
list_milestones / save_milestone read / write Milestones, their dates and their cases.

Manual runs — manual-runs:read, manual-runs:write

Tool Needs What it does
list_manual_runs read Recent runs, and how far each got.
get_manual_run read One run: its cases, their steps, and every result so far.
start_manual_run write Start a run of manual cases.
record_step_result write Record one step as passed, failed, blocked or skipped, with a note.
finish_manual_run write Finish it, turning step results into case results.

How it behaves

  • New cases are Drafts, marked as made by that token. A person marks them ready.
  • It searches before it creates. A case with the same name is refused, with the id to update instead.
  • Refusals teach. "This token cannot do that (it needs the cases:write scope)" is what the assistant reads, and repeats to you.
  • Everything it returns is labelled as data — case text and results are never treated as instructions.