Skip to main content

IDE Integration Overview

Verdictan supports two different IDE integration paths. They apply to different tasks, and you can use them together.

PathEndpointWhat it governs
Hosted MCPhttps://<published-hostname>/mcpVerdictan tools and resources invoked by the IDE agent
Model APIhttp://127.0.0.1:41002/v1 or a published model hostnameModel requests sent by a client that supports a custom API endpoint

MCP does not reroute an IDE's built-in model traffic. It adds tools such as documentation search, policy validation, history queries, and context-fabric operations. To govern prompts and model responses, the IDE or extension must also support a custom model endpoint. Prefer Chat Completions or Responses. Chat Completions, Responses, and Messages use the documented access, input, tool, output, and audit stages. Select the request family that the IDE and configured provider support.

Select the right path​

If the client supports remote streamable HTTP MCP servers, use hosted MCP. Cursor, VS Code, Windsurf, Zed, Codex, and Claude Code can send an MCP bearer header. JetBrains AI Assistant has a remote MCP path. Use it only when the installed version has a protected bearer-header field.

If the client exposes an OpenAI-compatible base URL or a custom-endpoint provider, use the model API. Active examples include Zed, Continue, JetBrains AI Assistant, VS Code custom endpoint models, and other compatible editor plugins.

Do not point HTTP_PROXY, HTTPS_PROXY, or an IDE's generic network proxy at port 41002. The Verdictan model endpoint is an LLM API. It is not a forward HTTP proxy. A proprietary assistant cannot route model traffic through this gateway unless it supports MCP or a custom model endpoint.

Authentication boundaries​

  • The gateway process uses its own runtime token in VERDICTAN_API_TOKEN.
  • MCP clients authenticate to a published hostname. They use an active Verdictan API token, shown in the console as an Access Key or Gateway Key.
  • Model clients send a different Verdictan API token as their bearer/API key.
  • Provider credentials stay on the gateway. policy-config.yaml references them. They are not model-client credentials.

The shared API-token system supplies Access Keys and Gateway Keys as filtered views. It does not use different credential stores for these keys.

Recorded data​

Requests that pass through the gateway generate policy and audit events for their configured flow. Spend data must have the applicable usage and price data. History content is available only when capture is enabled. Verdictan cannot record proprietary IDE traffic that continues to go directly to the vendor.

When you test, use verdictan events tail --follow. A matching event proves that the request traversed Verdictan. If no event is available, use that as diagnostic input. It is not proof of bypass.

Make sure that the gateway mode delivers events. Make sure that the CLI queries the same organization and region. Verify that the time window and filters include the request. Then verify that the active model base URL points to the gateway.

Per-client guidance​

Next steps​