Connect a coding agent
Vigilon runs an MCP server, so a coding agent can read your production errors, traces and job runs while it works in your repo. You sign in with your Vigilon account. There is no API key to create or paste.
Before you start
A Vigilon account in a workspace. The agent reads what you can read in the dashboard, so the account you sign in with must already belong to a workspace.
A supported client. Claude Code, Codex, VS Code with GitHub Copilot, Claude and ChatGPT. Other MCP clients are added as we verify them.
Add the server
Pick your client and add the server once. This saves the server and does not sign you in yet.
claude mcp add --transport http vigilon https://mcp.vigilon.io/mcpAdd --scope user to make the server available in every project instead of only the folder you ran the command from.
Sign in
Signing in opens the Vigilon sign-in page in your browser and then asks you to approve read access to telemetry. Sign in with the account you use for the Vigilon dashboard.
Claude Code. Run /mcp, choose vigilon and authenticate.
Codex. Run codex mcp login vigilon.
VS Code. Start the server from the MCP servers list. VS Code asks for permission to sign in; allow it.
Claude and ChatGPT. Choose Connect on the connector you added.
Ask about production
Ask in plain language. The agent picks the tools, reads the data and then works in your code.
- What started failing in production in the last day?
- Find the newest error group in checkout-api and show me the stack trace.
- Which background jobs are failing, and what error did the last failed run hit?
- Investigate the regressed error group in the billing project and fix it.
What your agent can read
Every tool is read-only and works inside your workspace. When you leave out the environment, a tool uses the project's default environment, and when you leave out the time range, it uses the last 24 hours.
| Tool | Returns |
|---|---|
| list_projects | The projects in your workspace. |
| get_project_overview | Request volume, error rate and latency for a project and each of its services, compared with the previous period. |
| list_error_groups | Error groups with their counts, first and last seen times, and whether each is resolved, regressed or muted. |
| get_error_group | One error group, the endpoints or jobs it affects, and its most recent occurrences. |
| get_error_group_sample | The exception type, message and stack trace of one occurrence. |
| search_traces | Request traces, filtered by endpoint, method, errors or duration. |
| get_trace | One trace with its spans, attributes and events. |
| list_failed_requests | Individual requests that ended in a 5xx response, each linked to its error group and trace. |
| list_jobs | Background jobs with run counts, failure rate, duration and health. |
| get_job_detail | One job, compared with the previous period. |
| list_job_runs | The runs of one job, with status, duration and error. |
Access and security
The agent acts as you. It sees your workspace only and has the permissions of your role.
It can read, not change. An agent cannot resolve or mute error groups, and it has no access to settings, users, API keys, or billing.
It never sees your password. You sign in on the Vigilon sign-in page in your browser, and the agent receives a token limited to reading telemetry.
You can disconnect at any time. Remove the server or connector from your client. In Claude Code:
claude mcp remove vigilonLimits
120 requests a minute for each user. Every agent you connect shares that allowance. Past it, the server answers 429 with a Retry-After header.
Up to 50 rows a page. Lists return 20 rows unless the agent asks for more, and the agent pages through longer results.
Large results are shortened. A stack trace longer than 20,000 characters and a trace with more than 200 spans are cut, and the result says so. Failed spans are kept first.
Troubleshooting
The browser shows “The Client ID Metadata Document could not be resolved.” Your MCP client, or its version, is not one Vigilon accepts yet. Use a supported client and update it to its latest version.
The sign-in succeeds, then the client says the server rejected the credentials. The account you signed in with does not belong to a Vigilon workspace. Sign in with the account you use for the dashboard, or ask a workspace owner to invite you.
A tool reports that the project has no default environment. Name the environment in your prompt, for example production, or set a default environment in the project's settings.
A list comes back empty. The default time range is the last 24 hours. Ask for a longer range, and check that the project receives telemetry in that environment.