Other agents and MCP

Use Argot skills and read-only MCP context with a compatible host, without assuming lifecycle automation.

Argot’s CLI is the complete, explicit changeset check. Agent integrations can help an agent choose that workflow or request context, but they do not create a universal automatic check.

Skills: invoked workflows

Install the skill bundle in a compatible host:

npx skills add get-tmonier/argot

The bundle contains argot-setup, argot-refresh, argot-check, argot-review-pr, argot-setup-ci, argot-write-rule, and argot-suggest-rules. argot-refresh re-audits scope, structural paths, and mutes before a deliberate fit. A user or agent must select a skill; installation does not configure a hook or MCP server, and a skill does not schedule Argot commands on its own. The selected workflow may require the argot CLI and a fitted repository.

Claude Code has a packaged plugin path; see Claude Code. Cursor and Codex can be compatible skills or MCP hosts, but this repository does not ship or test a Cursor- or Codex-specific Argot lifecycle configuration. Do not treat either as an automatically checked host.

MCP: client-invoked context

argot mcp is a local stdio Model Context Protocol server. Configure a client that supports MCP, then let that client decide when to call it:

{
  "mcpServers": {
    "argot": { "command": "argot", "args": ["mcp", "--repo", "."] }
  }
}

The server exposes six consistently named, read-only tools:

Tool	Use it when	Scope
`argot.get_fit_status`	Before trusting learned state	Snapshot completeness, fit suitability, compatibility, and adaptive refresh reasons.
`argot.get_voice_context`	Before writing a file	Familiar imports, typical callees, and active replacement guidance for that language.
`argot.check_hunk`	While drafting an isolated snippet	Fast fitted-voice signal for one supplied hunk; it does not run the other detector groups.
`argot.explain_hunk`	After `check_hunk` needs investigation	The same hunk-level voice result with untruncated identifiers and attestation evidence.
`argot.check_changeset`	After editing real repository code	The complete configured voice, semantic, architecture, integrity, and custom-rule pipeline over the worktree, index, range, or one commit.
`argot.list_conventions`	When exploring how the team organizes code	Learned internal APIs, placement concentrations, and migrations still in progress.

A fitted repository is required for model-dependent context and checks; the server returns a readable tool error when that prerequisite is missing. MCP is passive: connecting it does not cause a check before a commit, at the end of an agent turn, or at acceptance time. Even check_changeset does not update .argot/last-check.json.

Fitting is intentionally not an MCP tool. It changes the shared learned baseline and therefore belongs to the local argot-setup or argot-refresh workflow, where scope and mutes are reviewed before the resulting .argot/ diff is committed.

Recommended agent loop

Set up the repository with Init and Fit.
Optionally call argot.get_voice_context before generating a change.
Run argot.check_changeset or the full argot check on the intended changeset.
Read the rule evidence; keep the human decision explicit.
When fit status recommends maintenance, use argot-refresh; never fit in CI.

Copying Argot’s AGENTS.md into a repository can give agents the same interpretation contract, but it does not install or trigger Argot.