Ask your coding agent why production broke
Vigilon's hosted MCP server gives Claude Code, Codex, VS Code, Claude and ChatGPT read access to your error groups, traces and job runs, so your agent debugs from real production data. Included on every plan.
vigilon · list_error_groups→ SignatureVerificationError · 87 timesvigilon · get_trace 8dc4a12f→ fails at webhook.ts:84The failures all come from signature verification. Today’s deploy parses the request body before Stripe sees it, so the raw payload is gone. The change is in webhook.ts. Want me to patch it?
Your agent reads production, you decide what changes.
Errors with their stack
List error groups with their counts, first and last seen and state, then fetch one occurrence's type, message and stack.
Traces span by span
Search traces by endpoint, method, errors or duration, and read any trace's spans, attributes and events.
Jobs and their runs
Read each job's run count, failure rate, duration and health, and the status and error of each run.
Insights on Team and Pro
Read the anomalies Vigilon found, with their severity, current reading and the baseline they were compared against.
Read-only by design
Every tool reads and none writes. Your agent cannot resolve or mute errors, or reach settings, users, API keys or billing.
No API key to paste
Sign in with your Vigilon account in the browser and approve read access. The agent sees your workspace only.
Connect once, then just ask.
- 01
Add the server
Run one command in Claude Code or Codex, or add the server URL in VS Code, Claude or ChatGPT.
- 02
Sign in
Your browser opens the Vigilon sign-in page. Approve read access for your workspace.
- 03
Ask about production
Ask what started failing or which job missed its run. The agent picks the tools and reads the data.
- 04
Review the fix
The agent works in your code with what it found, and you decide what ships.
One command in your coding agent.
Add the hosted server, sign in in the browser, and ask. MCP queries do not count against your plan's quota.
claude mcp add --transport http vigilon https://mcp.vigilon.io/mcpThen run /mcp, choose vigilon and sign in. Add --scope user to use it in every project.
AI debugging in Vigilon and Sentry.
Sentry has its own MCP server and sells an AI agent on top. This is how the two compare, with a source for every Sentry fact.
- [3] Seer pricing: sentry.io/product/seer. Checked 2026-10-25.
- [4] Sentry MCP server: mcp.sentry.dev. Checked 2026-10-25.
MCP server questions.
What does Vigilon's MCP server do?
MCP, the Model Context Protocol, is how coding agents call outside tools. Vigilon's MCP server lets an agent read your production error groups, traces, job runs and insights while it works in your repo, so it can debug from what actually happened.
Why not paste the stack trace into the chat?
A pasted stack trace is one occurrence, copied by hand. Through the server the agent reads the whole error group: how often it happens, when it started, whether it came back after being resolved, which endpoints or jobs it hits, and the trace of a failing request. After you deploy a fix, it can check whether the error is still happening.
Which coding agents work with it?
Claude Code, Codex 0.151.0 or later, VS Code with GitHub Copilot, Claude and ChatGPT. Other MCP clients are added as they are verified.
Can the agent change anything in Vigilon?
No. Every tool is read-only and works inside your workspace with the permissions of your role. It cannot resolve or mute error groups, and it has no access to settings, users, API keys or billing.
Do I need an API key?
No. You sign in with your Vigilon account in the browser and approve read access; there is no key to create or paste.
Does it cost extra?
No. The MCP server is included on every plan, and its queries do not count against your quota. Reading insights needs Team or Pro.
Everything else Vigilon watches.
Error tracking
Repeated errors grouped into one, with an email when a new error appears or a fixed one comes back.
Cron job monitoring
Every run of a scheduled job or queue worker is a trace, and a job that misses its schedule shows as overdue.
Distributed tracing
Every request with an error and every slow request kept as a full trace, with its database queries and outgoing calls.
Endpoint monitoring
Traffic, p50 and p95 latency, error rate and status codes for every route, counted from all requests before any trace is dropped.
Anomaly detection
Error spikes, slower endpoints, error surges, missed job runs and failing job streaks, each checked against its own history. On Team and Pro.
Start free for two months.
Pick a plan, add the SDK, and Vigilon starts watching your app. Starter is free for your first two months, with no card needed.