What it does
The server gives AI agents tools to search active listings by keyword, category, university and maximum price, fetch full listing details, read public market statistics, list valid categories, create draft listings and send messages to a listing's poster. A message is only submitted: a confirmation link is emailed to the sender and the message goes out only after that link is clicked, so agents report it as pending. A status tool reports whether the confirmation email was delivered. No personal information is returned. The same codebase also runs for SUpost, selected by a brand setting.
An agent would reach for it to browse or search the marketplace, check market figures, or prepare a listing for the poster to finish. It is a pure client of public surfaces: it holds no secrets and has no privileged database access, and its only configuration is the public base URL. It runs as a stateless streamable-HTTP endpoint on Vercel with no sessions and no authentication, and a health route reports the deployed commit version.
Running it means connecting an MCP client to the deployed endpoint. Optional environment variables override the base URL for preview deployments and tune or disable delivery-status polling. Uptime monitoring of the health route is a separate manual setup step.
Limitations and requirements
create_post only creates a draft; the poster must add photos, review and publish, and pay first when not on the free tier. Messages are delivered only after the recipient clicks the confirmation link. An address whose confirmation email hard-bounced in the last 30 days is refused with an error. The status check reports only the confirmation email's delivery, never whether the link was clicked. The health route does not confirm that the upstream site is up. Production deploys come from the GitHub integration on master; a local vercel deploy --prod can overwrite the production alias if the checkout is stale.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
io.github.Capmus-Team/capmus-mcp
Search listings, message posters, and create draft posts on Capmus, the university marketplace
v0.2.2As shown on registry.modelcontextprotocol.io when we last read it