Guides

Which coding agents can read my production errors?

Any coding agent that supports MCP servers can read production errors from an error tracker that has one. 14 clients connect to Runheed with a URL and a browser sign-in, among them Claude Code, Cursor, Codex, VS Code, Gemini CLI and ChatGPT. Windsurf and Cline use the local server with an agent key.

Every agent with a setup page

AgentHow it connectsSetup
claude.aiServer URL and a browser sign-in, with no key to copy.claude.ai setup
ChatGPTServer URL and a browser sign-in, with no key to copy.ChatGPT setup
Claude CodeServer URL and a browser sign-in, with no key to copy.Claude Code setup
CursorServer URL and a browser sign-in, with no key to copy.Cursor setup
Claude DesktopServer URL and a browser sign-in, with no key to copy.Claude Desktop setup
VS CodeServer URL and a browser sign-in, with no key to copy.VS Code setup
WindsurfLocal server (runs with npx) and an agent key from Settings > Agent keys.Windsurf setup
Codex CLIServer URL and a browser sign-in, with no key to copy.Codex CLI setup
Gemini CLIServer URL and a browser sign-in, with no key to copy.Gemini CLI setup
KiroServer URL and a browser sign-in, with no key to copy.Kiro setup
LovableServer URL and a browser sign-in, with no key to copy.Lovable setup
ReplitServer URL and a browser sign-in, with no key to copy.Replit setup
BoltServer URL and a browser sign-in, with no key to copy.Bolt setup
AntigravityServer URL and a browser sign-in, with no key to copy.Antigravity setup
ZedServer URL and a browser sign-in, with no key to copy.Zed setup
ClineLocal server (runs with npx) and an agent key from Settings > Agent keys.Cline setup

Each setup page has the config file or command for that client, checked against the client's own MCP docs, and a prompt that confirms the connection works.

What the agent can do once connected

  • List recent errors, search them, and read one in full: stack trace, request, breadcrumbs, environment and release.
  • See trends over 24 hours, 7 days and 30 days, and a health score per project.
  • Resolve an error with a note such as the commit hash, ignore known noise, or merge duplicates. A resolved error that happens again reopens.
  • Create a project and send a test event, so it can add reporting to your app and check that errors arrive.

Every client gets the same thirteen tools. The MCP server page lists them by name.

URL or local server?

Use the URL wherever the client supports it. Nothing goes into your repository, and the connection appears in Runheed under Settings > Agent keys > Connected apps, where Disconnect ends it. Choose Read only on the consent page if the agent should only read.

Use the local server where nobody can sign in through a browser, such as a CI runner, or in a client that only runs local servers. It needs Node.js 18 or newer and an agent key. Keep the key in an environment variable or a secret store, never in a committed file.

If your agent is not on the list

Check whether it supports MCP servers over Streamable HTTP. If it does, give it https://runheed.com/api/mcp and sign in when it asks. If it only runs local servers, copy the mcpServers block from the Claude Desktop page; most clients use that format. For Lovable, Replit and Bolt, see error tracking for Lovable, Replit and Bolt apps.

Questions

Can GitHub Copilot read production errors?
Yes, in VS Code agent mode. VS Code runs MCP servers for Copilot: add the Runheed URL to .vscode/mcp.json and sign in when VS Code asks.
Can ChatGPT or claude.ai read them too?
Yes. Runheed is listed in ChatGPT's plugin directory and Claude's connector directory. Both connect by URL with a sign-in, which is handy when you triage away from the editor.
Do I need a separate account for each agent?
No. Connect several clients to the same workspace; each one shows up as its own connected app.
What does it cost?
Plans start at $29/month for 50,000 events a month, 3 projects and 30 days of history.

Related