Sentry MCP: investigate a production error and propose a fix

Sentry MCP lets an AI coding agent retrieve error evidence from Sentry and use it alongside your code to investigate a production failure. The official Sentry MCP server offers a remote HTTP connection with OAuth sign-in. Use it to support a proposed fix, then reproduce the failure and review the patch before merging. Sentry connection documentation

What it can do

Sentry's documented investigation workflow includes issue details, stack traces, breadcrumbs, event filtering and related traces where available. These give an agent evidence to compare with your repository. Official debugging workflow

Give the agent a bounded question: inspect this issue, identify the failing code path and explain what evidence would distinguish the likely causes. Keep investigation separate from permission to change code or issue status. For the wider terminology, see what MCP is; related connections are grouped under MCP servers for apps.

A useful answer separates observation from inference. An exception identifies where execution failed; an explanation of why it failed still needs checking against code and a reproduction. Ask for a source link for each observation and an explicit list of missing evidence.

From a production error to a reviewable fix

Consider a hypothetical checkout failure after a deployment. An event suggests that a delivery address was absent when the application tried to read its postcode. That is a starting hypothesis, not a reason to insert a default value immediately.

First, ask the agent to relate the error to the affected code and deployment context. Have it explain whether an absent address is allowed by the application contract. A customer collecting an order might legitimately have no delivery address; silently inventing one could hide a business logic error.

Next, reproduce the relevant condition with synthetic data in an isolated working copy. Write a regression test that fails before the patch. If the agent cannot reproduce the symptom, retain the uncertainty and request the missing context instead of presenting a speculative change as verified.

Then make the smallest justified change. Run the regression test and the relevant existing checks. Reviewers should receive the issue link, a concise hypothesis, the code diff, commands actually run and their results. State which behaviours remain untested. Merge and deployment remain separate decisions; decide how the affected path will be observed afterwards.

A practical investigation prompt is:

Investigate the supplied Sentry issue in the selected project.
Separate event evidence from hypotheses. Compare the evidence with
this repository, propose a synthetic reproduction and a minimal fix.
Return the diff and observed test results for review.
Do not change issue status, merge or deploy.

This is a suggested task boundary, not a technical permission control. Configure the client's tools and approvals to match it.

How to connect it safely

1. Choose the project and the client. Use a client that supports remote HTTP MCP and OAuth. Sentry documents project-scoped URLs and recommends project scoping where possible. Replace the placeholders in https://mcp.sentry.dev/mcp/{organizationSlug}/{projectSlug} with your actual slugs. Endpoint and scoping

2. Add the remote endpoint in VS Code. Run MCP: Add Server from the Command Palette, select HTTP and enter the URL. Review the configuration destination and trust prompt. Follow the current VS Code instructions and the editor walkthrough.

3. Complete OAuth deliberately. Check the account, project restriction and requested permissions before consenting. Sentry documents separate MCP and upstream Sentry authorization layers; both constrain access. Do not paste tokens into a conversation or commit them with configuration. OAuth architecture

4. Verify a low-impact request. Request a known, sanitized issue without mutations. Inspect the actual tool call and returned evidence. Keep approval for consequential actions. VS Code verification guidance

5. Grant local actions separately. Allow repository inspection, test execution and file edits only within the intended working copy. Keep merging, deployments and production changes behind explicit review. Use the MCP security checklist before broadening access.

What data the agent may see

Treat a production event as potentially sensitive. Sentry's workflow warns that messages, request data and user context may contain credentials or personal information. It also treats event content as untrusted: a command embedded in an error message is evidence, not an instruction. Sentry data-handling guidance

Before the first production query, decide whether the returned fields may enter your assistant's processing and conversation history. Review that client's retention and processing terms. OAuth controls access; it does not establish your organization's permission to disclose the resulting content to another service.

Sentry documents project settings for removing sensitive values and applying additional scrubbing rules before event storage. Review those settings, inspect representative sanitized output, and avoid copying raw payloads into patches or reports. Project data-scrubbing settings For the broader hosting decision, compare remote and local MCP servers.

When this does not work

The evidence cannot support a reproduction. If the returned context does not explain the failure, ask for a clearer example or relevant application evidence. Do not convert missing information into confidence. Verify that the checkout matches the affected code before editing.

The connection needs reauthorization. Sentry's OAuth implementation can require a fresh sign-in when cached authorization is no longer usable. Check the connection and approved scope before changing permissions. Authorization lifecycle

Requests are throttled. Sentry applies API limits to request frequency and concurrency. Reduce parallel investigation calls and follow the returned retry guidance. API rate limits

You operate self-hosted Sentry. The official repository describes a local stdio route, still a work in progress, with separate token configuration. Some AI search tools require additional provider configuration. Check the current documentation for your deployment rather than assuming hosted behaviour. Official repository

Where Tvijo fits

The intended sequence is error → reproduce → fix → receipt, with review before merge. Sentry supplies diagnostic context; the coding client needs its own repository and execution permissions. Tvijo's role is a bounded task and review workflow, not an advertised ready-made Sentry integration.

As of 30 Oct 2026, the DSHREM VS Code extension provides an editor entry for selected context and task receipts. Use sanitized diagnostic context with the relevant code and define acceptance checks. TvijoCode is documented as a test build with the extension embedded; the described build is Windows-only, unsigned and without automatic updates.

Tvijo MCP makes Tvijo callable by a compatible MCP client. Its published consultation path does not automatically make the Tvijo editor a Sentry MCP client. Supplied-diff review is prepared in merged source, with package publication pending; editor integration and safe delegation are planned stages. The prepared review does not run tests, edit the repository or merge changes.

Keep the task receipt alongside the patch and observed test results. Spending evidence does not prove the bug is fixed. A reviewer still needs to assess the change and its validation.

As of 30 Oct 2026, Tvijo's offers show Try at $0: up to 5 tasks, 50,000 tokens or $0.05 of sponsored usage, whichever is reached first, from a shared $1 daily pool funded by TVIJO. Service credit is shown as $10 for requested prepaid credit, with metered provider usage plus an 8% service markup. Payment collection is not activated; an order is not a charge.

FAQ

Can an AI agent fix a production error through Sentry MCP?

It can use diagnostic evidence to propose a change. A supported fix still needs repository access, reproduction, tests and human review; MCP access alone does not establish correctness.

Do I need to run the Sentry MCP server locally?

The documented hosted route uses a remote HTTP endpoint and OAuth. Self-hosted deployments have a separate stdio option; check the repository instructions.

Does connecting Sentry authorize an automatic merge?

No. Keep merge permission separate and review the diff, observed test results and remaining uncertainty before accepting the change.

Sentry is a trademark of its owner. Tvijo is not affiliated with or endorsed by it.