How to connect Vercel
stromboli reads a Vercel project’s runtime logs and deployments through Vercel’s MCP server.
1. Log in
Vercel’s MCP server takes no access token: stromboli logs in with OAuth, once, in your browser. Run this in a terminal, with your team’s slug (as in vercel.com/<slug>) or ID:
stromboli connector add vercel vercel team_id=<team-slug>
It opens Vercel’s consent page; approve it, and the connector is saved and tested. The test checks that the login can read the team’s projects, so a wrong team_id fails here. If the browser runs on another machine (over SSH, for example), open the printed URL there and paste back the address it redirects to.
The login has the access of your Vercel user; stromboli may only call list_projects, list_deployments and get_runtime_logs. The server keeps the connector working on its own. If the login expires or you revoke it in Vercel, calls fail with run stromboli connector login vercel: run it to log in again.
2. Add the services
Let the connector propose one service per project, with its logs and deploys filled in:
stromboli catalog discover vercel --owner @you
Found 3 services: api, docs, web.
Add them to the catalog? [Y/n]
Added 3 services to the catalog.
Each service comes out like this:
- name: web
owners: ["@you"]
logs:
source: vercel
selector: web # project name or ID
deploys:
source: vercel
project: web # project name or ID
Add what each project depends on as you go, e.g. stromboli catalog set web --add-depends-on api. To narrow it down, redirect discover to a file (> vercel.yaml), edit the entries, then run stromboli catalog import vercel.yaml:
logs:
selector: web environment=production # environment is optional
deploys:
environment: production # optional: production (default) or preview
branch: main # optional: only deploys of this branch
3. Check it
stromboli catalog validate --check-connectors
stromboli brief web --since 1h
The Logs section shows the error patterns of the project’s runtime logs, and Recent changes its deployments with their commit, author and state (ready, error…).
Limits
- Vercel’s MCP server returns log entries that carry a message: what your functions print, and errors. Plain request records (the
GET 200 /rows without a message in the dashboard) are not included, so a project that serves only static or cached pages shows 0 lines, and 5xx responses without a logged error are not seen. To get request records too, send the logs to Axiom (below). - Vercel returns at most 100 log lines per call. Counts without a filter are exact; counts of error-like lines stop at 500.
- Logs go back as far as your Vercel plan keeps them. For longer history, also send them to Axiom with the Axiom integration and connect Axiom instead.