Connect a Jenkins MCP Server to Claude Desktop, Cursor and VS Code

A Jenkins MCP server can expose approved build information and carefully controlled actions to compatible AI clients. The client connection is only one layer: the server must still enforce authentication, allowlists, input validation, redaction, rate limits, and audit logging.

Choose the transport model#

Prerequisites#

• The Jenkins MCP server starts successfully outside the AI client.
• Read-only tools have been tested with synthetic or non-production jobs.
• Jenkins uses a dedicated least-privilege service account.
• Secrets are stored outside source code and client-visible prompts.
• The client version supports the server’s transport and MCP capabilities.
• Server logs include correlation identifiers without exposing credentials.

Client configuration pattern#

Claude Desktop, Cursor, and VS Code each maintain their own MCP configuration format and location. For a local server, the entry normally identifies an executable, its arguments, and required environment references. For a remote server, it identifies an HTTPS endpoint and the supported authentication method. Exact fields can change between client releases, so validate against current client documentation.

Connect one client at a time#

1. Start the MCP server manually and confirm its health.
2. Add one minimal client entry without production credentials.
3. Restart or reload the client as its documentation requires.
4. Confirm the server appears and exposes only the expected tools.
5. Run a harmless read-only request such as approved job status.
6. Review both client and server logs for the same correlation ID.
7. Repeat for the next client only after the first connection is stable.

Protect credentials#

Do not put Jenkins administrator credentials in desktop configuration. Prefer a server-side secret store and a dedicated Jenkins API token scoped to approved jobs. Treat client environment variables as accessible to the local user and potentially to extensions running in that environment.

Troubleshooting#

Keep state-changing tools disabled until every client passes read-only validation. After enabling an approved action, require explicit target parameters, environment boundaries, and human confirmation for sensitive jobs.

Model Context Protocol documentation