MCP for Self-Hosted Error Tracking

Why MCP matters for error tracking

Error tracking is full of context: stack traces, culprit frames, request data, breadcrumbs, users, environments, releases, notes, assignments, and the history of what your team has already tried. That context is valuable, but it often sits one browser tab away from the code editor where the fix actually happens.

MCP (Model Context Protocol) gives AI coding tools a standard way to ask Telebugs for that context. Instead of pasting a stack trace into a chat window, you can connect Claude, Cursor, Windsurf, OpenAI/Codex, or another MCP client to your Telebugs instance and let it inspect the same facts you see in the dashboard.

The goal is not magic. The goal is a shorter path from "production is broken" to "this is the file and line I need to change."

What an AI coding tool can see

Telebugs exposes a Streamable HTTP MCP endpoint at /mcp. After authorization, an MCP client can use Telebugs tools to read the debugging context that matters:

• List projects, project users, error groups, reports, and notes.
• Inspect a specific error group or report, including backtraces and breadcrumbs.
• Filter error groups by status, query, date range, and severity.
• Resolve, unresolve, mute, unmute, assign, unassign, merge, and add notes when write access is granted.
• Run bulk resolve, bulk mute, and bulk merge actions for cleanup workflows.

That means your coding tool can start from live production evidence: "show me the newest unresolved Rails errors," "inspect the top Next.js exception from the last deploy," or "summarize the breadcrumbs for this failing checkout report."

Secure by default, not a secret backdoor

MCP access should feel like a normal connected app, not a hidden admin tunnel. Telebugs uses OAuth with PKCE, dynamic client registration, short-lived access tokens, refresh tokens, and revocation.

Clients request explicit scopes:

• telebugs.read for reading projects, errors, reports, notes, and context.
• telebugs.write for actions like resolving, muting, assigning, merging, and adding notes.

Connected MCP apps can be reviewed and revoked from account settings. If a tool is no longer trusted, remove its access without digging through the database.

Privacy-friendly AI debugging

Telebugs is self-hosted, so your production error tracker lives on your infrastructure. MCP does not change that. A coding tool connects to your Telebugs instance with your permission, using the scopes you approved.

Telebugs also marks values from error reports as untrusted in MCP responses. Error messages, breadcrumbs, frame context, request values, and similar fields can contain user-provided input. AI tools should use that data for debugging, not treat it as instructions.

This is especially useful for teams that want AI-assisted debugging but do not want to move error tracking into another hosted observability product. See the privacy-first error tracking guide for the broader data-control story.

Where this fits

MCP is a practical layer on top of the core Telebugs workflow:

• Error tracking stays first. Telebugs still collects, groups, and notifies you about production errors.
• Sentry SDK compatibility still matters. Your apps keep sending rich error context through familiar SDKs.
• Notifications still wake the right person. MCP helps after the alert, when someone needs to inspect and fix the issue.
• AI is optional. Telebugs remains useful without any AI tool connected.

For the release story and implementation details, read Telebugs 1.17.0: Introducing MCP Support.

Frequently asked questions

What is MCP?

MCP is the Model Context Protocol, a standard way for AI tools to connect to external systems and request structured context or actions. Telebugs uses it to expose error tracking data to authorized MCP clients.

Which AI coding tools work with Telebugs MCP?

Telebugs is designed for MCP clients such as Claude, Cursor, Windsurf, OpenAI/Codex, and other tools that support HTTP MCP servers with OAuth authorization.

Can an AI tool change my errors?

Only if you grant write access. Read-only access lets the client inspect projects, groups, reports, breadcrumbs, backtraces, and notes. Write access is required for actions like resolving, muting, assigning, merging, or adding notes.

Does MCP send my error data to Telebugs cloud?

No. Telebugs is self-hosted. The MCP endpoint runs on your Telebugs instance, and connected clients access that instance with the permissions you approve.

Can I revoke MCP access?

Yes. Connected MCP apps can be reviewed and revoked from account settings.

Ready to debug production errors from your coding tool?

Read the MCP release notes, browse the Telebugs manual, or get Telebugs and connect an MCP client to your own error tracker.