Skip to main content

JetBrains AI Assistant with Verdictan

JetBrains AI Assistant supports OpenAI-compatible endpoints. It also supports custom MCP servers, but the two paths use different authentication settings.

Route model requests

  1. Start the gateway using Gateway Setup for IDEs.
  2. Open Settings → Tools → AI Assistant → Providers & API keys.
  3. Add an OpenAI-compatible endpoint.
  4. Set the endpoint to http://127.0.0.1:41002/v1.
  5. Add a Verdictan client API token.
  6. Test the connection.
  7. Assign a returned model to the AI Assistant feature you intend to use.

JetBrains assigns custom models to feature groups. A feature can be unavailable when the selected model does not have a necessary protocol or capability. Inline completion also has model-specific requirements. Verify that a chat model supports fill-in-the-middle completion before you assign it.

See JetBrains' custom-model documentation for settings and assignment rules.

MCP authentication requirement

The Verdictan MCP endpoint must have an HTTP bearer header. At this time, the public JetBrains MCP documentation documents remote server URLs. It does not document a protected bearer-header field for that remote configuration.

Do not put a raw Verdictan token in a project MCP file. Add the hosted MCP endpoint only if your installed JetBrains version has a protected bearer header field. Verify that the field sends the header to https://<published-hostname>/mcp. MCP tools supplement the selected model. They do not reroute a JetBrains-hosted model.

Verify

Run verdictan events tail --since 10m --follow. Select the Verdictan model. Send a chat request. A matching event verifies the model path. If you configured MCP with a protected header, examine its server status and test a tool separately.

Next steps