Compare / MCP servers
Error tracking MCP servers compared
Every error tracker below has an MCP server, so Claude Code, Cursor, claude.ai or ChatGPT can read your production errors. They differ in where the server runs, what your agent can do with an error, and how much context comes with it. Checked 2026-11-02 against each vendor's own pages.
| Tool | MCP server | Agent context on errors | Entry paid plan |
|---|---|---|---|
| Runheed | Yes, hosted MCP server at runheed.com/api/mcp (OAuth), plus a local stdio server | Agent name, session and tool are fields on every error, returned by the MCP tools | Solo $29/mo with 50,000 events, 30-day history |
| Sentry | Yes, hosted MCP server at mcp.sentry.dev (OAuth) | Custom tags and contexts you set yourself | Team plan $26/mo billed annually ($29 monthly) with 50k errors and up to 90-day lookback |
| Honeybadger | Yes, hosted MCP server (or run it yourself with Docker) | Custom context you set yourself | Team plan $26/mo with 50K errors and 90-day error retention |
| Rollbar | Yes, open-source (MIT) local stdio MCP server | Custom data you set yourself | Essentials from $9/mo for 10,000 occurrences, 90-day retention |
| BugSnag | Yes, SmartBear's self-hosted MCP server (npm package @smartbear/mcp) | Metadata you attach yourself with addMetadata or an onError callback | Select plan, priced by event tier from 50k events a month, 60-day retention |
Sources: Sentry MCP, Honeybadger MCP, Rollbar MCP, BugSnag MCP. Pricing sources are on each comparison page.
What to check before you connect one
- Does it run where your agent runs?
- A hosted (remote) server works from claude.ai, ChatGPT and Claude Code with just a URL. A local stdio server needs Node or Docker on the machine and only works in desktop and CLI agents.
- Can the agent close the loop?
- Reading errors is half the job. Check whether the agent can mark an error resolved or ignored, so the next session does not chase the same bug.
- What context comes back?
- A stack trace says where it broke. Your agent fixes faster when the error also says which agent session and tool wrote or ran the code. Most trackers store this only if you attach it yourself.
- How does sign-in work?
- OAuth means you approve the connection in a browser and can revoke it later. Key-based servers need a token pasted into a config file.
Which one fits
- You need tracing, session replay or mobile SDKs in the same tool: Sentry or BugSnag.
- You want uptime checks and status pages on the same bill: Honeybadger.
- You are debugging prompts and model calls rather than exceptions: an LLM tracing tool such as LangSmith.
- Your coding agents write and ship the code, and you want them to find, fix and resolve their own errors with the session that caused them: Runheed.
Connect Runheed to your agent
One URL, sign in, nine tools. Listed in Claude's connector directory.