Install AgentNoah in your IDE

AgentNoah uses MCP (Model Context Protocol) to plug into your IDE. Think of MCP as a USB port — you paste a small config snippet, and your IDE gets new tools it can use.

All snippets need YOUR API key (from agentnoah.dev → Settings → Rotate Key after sign-up). The daemon URL is the same for everyone: https://agentnoah.dev/mcp

Path

~/.claude/mcp.json (macOS / Linux) — or %USERPROFILE%\.claude\mcp.json (Windows)

{
  "mcpServers": {
    "agentnoah": {
      "type": "http",
      "url": "https://agentnoah.dev/mcp",
      "headers": {
        "X-API-Key": "<paste your full API key here>"
      }
    }
  }
}

After paste: Restart Claude Code. Test by typing: "health_score for owner/repo"

Quick test after install

Restart your IDE, then type one of these in chat to confirm AgentNoah is wired up:

What's the health score for owner/repo?
Run an AgentNoah audit on this repo
@agentnoah audit my repo (VS Code Copilot syntax)

If your IDE's AI calls an AgentNoah tool (you'll see a JSON response or a status message), the install worked. If it doesn't recognize the tools, see the troubleshooting table below.

Known limitations

start_byol_audit files=[...] parameter is advisory. Right now this MCP tool's explicit file list is silently ignored — the audit brief returns the daemon's default file sample regardless of what you pass. Workaround for targeting specific files: have your IDE LLM fetch the files directly (e.g. WebFetch from raw.githubusercontent.com) and analyze them. The submit_byol_findings and finalize_byol_audit_run calls work correctly; findings persist against the audit_id. We're addressing this in the next release. The full-repo audit(repo="owner/repo") tool is not affected — it uses default file selection by design.

Troubleshooting

Symptom	Fix
IDE says "AgentNoah not found" or "MCP server unavailable"	Restart your IDE after pasting config. Some IDEs only re-read MCP config on launch.
Tools show up but every call returns "403 Forbidden"	Your API key is missing or wrong. Go to agentnoah.dev → Settings → Rotate Key, copy the full key (starts with ak_), paste over <paste your full API key here>.
Audit takes longer than 10 minutes	Repo might be very large. Try the "quick" pipeline (lighter scope, smaller file set) by asking your IDE: "Run an AgentNoah quick audit on owner/repo".
"Repo not connected to your account"	You haven't added the repo on agentnoah.dev/dashboard/settings/repos yet. Add it first, then try the audit.
Build PR gets opened but CI fails	Normal in some cases — the 16-step methodology's DEBUG step will fire automatically and propose a fix in a follow-up commit.

You're installed. Now what?

The walkthroughs page shows what a typical first-audit and a full BUILD session actually look like, step by step.

See the walkthroughs →