@still-running/mcp

An MCP server that puts stillrunning.dev’s workflow-analysis engine inside the tool you’re already using to build or debug an n8n workflow or Make scenario — Claude Desktop, Claude Code, or Cursor.

The analysis itself is @still-running/health-check, pinned to an exact published version: static analysis of the exported workflow JSON, no execution, no network call, no storage. Same zero-network rule as the still-running CLI that package ships — this server just puts an MCP tool interface on top of the same engine, plus a couple of tools (explain_finding, compare_workflows, list_rules) that only make sense in an interactive, conversational tool.

Tools

analyze_workflow

Checks an n8n workflow or Make blueprint export for silent-failure risks. Takes workflow (a JSON string or an already-parsed object — either the whole exported file). Returns the full AnalysisResult: findings, what’s protecting the workflow already, and a few counters.

explain_finding

Explains one of the four check types in general — what it looks for, why it matters, its severity range, known limitations. Takes checkId (the same string every finding from analyze_workflow carries: zero-write, credential-expiry, no-cadence, or error-handling). A finding’s own ifItGoesQuiet/detail/howToCheck fields already explain that specific instance — this is for the check’s general methodology instead.

compare_workflows

Runs analyze_workflow on two versions of the same workflow (e.g. before/after a fix) and reports which findings were added, removed, unchanged, or changed severity. Takes before and after, same shape as analyze_workflow’s workflow argument. Findings are matched by check + node — documented as a heuristic in the tool’s own output, not oversold as a perfect diff.

list_rules

Lists every check and every “protection” (the positive-signal side of the same analysis) the engine recognizes, with no input needed. Useful context to pull once at the start of a conversation rather than calling explain_finding four times.

Setup

Requires Node.js 18 or later. Every client below runs the server the same way — npx -y @still-running/mcp — Claude Desktop, Claude Code, and Cursor each just need to be told to run that command over stdio.

Claude Desktop

Edit your config file — quit Claude Desktop fully and reopen it afterward, a reload isn’t enough:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "still-running": {
      "command": "npx",
      "args": ["-y", "@still-running/mcp"]
    }
  }
}

Claude Code

From the CLI, in a project you want it available in:

claude mcp add still-running -- npx -y @still-running/mcp

To make it available in every project instead of just this one, add --scope user:

claude mcp add --scope user still-running -- npx -y @still-running/mcp

Or edit the JSON directly — .mcp.json in a project root (commit it to share with your team), or ~/.claude.json for user scope:

{
  "mcpServers": {
    "still-running": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@still-running/mcp"]
    }
  }
}

Verify with claude mcp list.

Cursor

Global (all projects), ~/.cursor/mcp.json — or per-project, .cursor/mcp.json in the project root:

{
  "mcpServers": {
    "still-running": {
      "command": "npx",
      "args": ["-y", "@still-running/mcp"]
    }
  }
}

Local development

npm install
npm run build
npm run typecheck
npm test

npm run build compiles src/ to dist/ — the published package ships only dist/ and this README (see files in package.json).

Resources

• stillrunning.dev
• @still-running/health-check — the analysis engine this server wraps
• Model Context Protocol