MCP Tools Reference

@clarxai/mcp exposes up to five tools. The first three always register; the AI tools register only when ANTHROPIC_API_KEY is set in the server environment.

analyze_manifest

Scores manifest content with the same rules as the Clarx Cloud manifest studio — identical input produces identical scores.

Input

Output — overall_score (0–100), score_type: "manifest_quality_estimate", pillar_scores (Discoverability, Boundary clarity, Context efficiency, Operational guidance, Edit safety), findings (rule, severity, line, description, section_key), sections checklist, and the outline.

analyze_clarx_manifest

Validates a clarx-manifest.json machine manifest.

Input — content or file_path (one required, as above).

Output

{
  "valid": true,
  "errors": [],
  "keys_recognized": ["generated", "verificationCommands"],
  "suggestions": [
    "Consider adding \"highFanIn\": Files many modules import — agents edit these with extra care."
  ]
}

Unknown keys and malformed thresholds overrides are reported; malformed overrides are ignored by the engine and can never disable a rule.

get_ci_workflow

Generates the GitHub Actions workflow described in CI Integration.

Input

Output — { "filename": ".github/workflows/clarx.yml", "yaml": "…" }. The agent writes the file; commit it and set up branch protection.

suggest_manifest_fix

Writes one well-formed markdown section to close an analyze_manifest finding. Runs against your ANTHROPIC_API_KEY (bring-your-own-key; requests go directly from your machine to the Anthropic API).

Input — section_key (e.g. build, testing, architecture — any section from the analyzer checklist), current_content or file_path, optional doc_path, repo_name, repo_description, repo_readme for grounding.

Output — { "section_markdown": "## Build\n…", "section_key": "build", "pillar": "O" }. The generated section satisfies the analyzer's validator rules (minimum words, required code blocks, required structure) so the finding actually closes.

generate_manifest_draft

First-draft manifest grounded in the repository README and description.

Input — filename (CLAUDE.md | AGENTS.md | GEMINI.md), plus repo_readme (preferred) and/or repo_description, optional repo_name.

Output — { "filename": "AGENTS.md", "markdown": "…" }. Claims the model cannot ground in the provided context are marked as <!-- TODO: … --> comments — resolve them before committing.

Hosted tools (Clarx Cloud token)

Create an API token in your Clarx Cloud org settings and set it as CLARX_MCP_TOKEN in the server environment. Four additional tools register, all org-scoped and enforced server-side:

list_repos

No input. Returns repo_id, full_name, latest_score, latest_scan_at, threshold per repository the token can access.

get_scan

Input — repo_id or repo_full_name (both accepted everywhere below); optional scan_id for the full detail of one scan.

Output — latest scan summary (overall score, pillar scores, confidence) with score_type: "repo_ai_readiness".

list_findings

Input — repo reference, optional severity array and pillar filters, limit (default 50, max 200), compact (default true — id/rule/severity/title/path only, so results don't crowd out your agent's context).

Output — findings, top_affected_modules, recommended_actions (whose ids feed get_remediation_prompt), and review_status.

get_remediation_prompt

Input — repo reference; action_id from recommended_actions (defaults to the top action); format: "prompt" | "brief"; combine_all: true to merge every action into one prompt.

Output — { prompt, action, working_set_files }. The prompt is identical to what the dashboard generates for the same scan — hand it to your coding agent and it will scope the patch to the working set.