@clarxai/mcp

@clarxai/mcp

Clarx MCP server — score agent manifests (CLAUDE.md / AGENTS.md), validate clarx-manifest.json, and generate Clarx CI workflows directly from your coding agent (Cursor, Claude Code, Claude Desktop, Windsurf, …).

No account required. The core tools run entirely on your machine with the same rules as the Clarx manifest studio.

Setup

Add to your MCP client config (e.g. .cursor/mcp.json, claude_desktop_config.json):

{
  "mcpServers": {
    "clarx": {
      "command": "npx",
      "args": ["-y", "@clarxai/mcp@latest"]
    }
  }
}

Claude Code:

claude mcp add clarx -- npx -y @clarxai/mcp@latest

Tools

Tool	What it does	Requires
analyze_manifest	Score a CLAUDE.md / AGENTS.md (0–100, pillar scores, findings with line numbers)	—
analyze_clarx_manifest	Validate clarx-manifest.json against the engine schema	—
get_ci_workflow	Generate a GitHub Actions workflow that gates PRs on the Clarx score	—
suggest_manifest_fix	AI-write one manifest section to close a finding	ANTHROPIC_API_KEY
generate_manifest_draft	AI-draft a full manifest from your README	ANTHROPIC_API_KEY

The AI tools are bring-your-own-key: set ANTHROPIC_API_KEY in the server env and they register automatically; without it they stay out of your agent's context.

analyze_manifest scores are manifest quality estimates — the same rules the manifest studio uses. They are not comparable to Clarx repo AI-readiness scores, which come from a full engine scan.

Hosted tools (Clarx Cloud token)

Create an API token in your Clarx Cloud org settings and set it as CLARX_MCP_TOKEN to unlock:

Tool	What it does
list_repos	Repositories the token can access, with latest AI-readiness scores
get_scan	Latest scan summary, or full drill-down by scan_id
list_findings	Findings + recommended actions from the latest scan (compact by default)
get_remediation_prompt	The structured fix prompt behind "Copy AI prompt" — feed it straight to your agent (format: brief for the shorter fix brief, combine_all for everything at once)

      "env": {
        "CLARX_MCP_TOKEN": "clarx_…"
      }

Tokens are org-scoped with scoped permissions; the API enforces plan limits server-side.