Connect an IDE to Verdictan: Model Traffic and MCP Tools

An editor can connect to Verdictan in two different ways. You can use one connection or both. This article helps you to select the connection and to verify it. IDE Integration Overview links to the guide for each editor.

Compare the two connections

Connection	Endpoint	What Verdictan governs
Model traffic	http://127.0.0.1:41002/v1, or your published model hostname	The model requests that the editor sends to this base URL
MCP tools	https://<published-hostname>/mcp	The Verdictan tool calls and resource calls from the agent of the editor

Only the model connection governs prompts and answers. For this connection, the editor must have a setting for a custom OpenAI-compatible base URL. The MCP connection does not change where the editor sends model requests.

warning

Do not set HTTP_PROXY, HTTPS_PROXY, or the network proxy setting of an editor to port 41002. That port serves an LLM API. It is not a forward HTTP proxy.

Get the three editor values

Each editor asks for the same three values:

Editor setting	Value
OpenAI-compatible base URL	The gateway base URL, including /v1
API key	The Verdictan application token
Model	A model id that GET /v1/models returns through the gateway

The API key field of the editor takes the Verdictan token. The provider key stays in the gateway. Gateway Setup for IDEs shows how to get and test the three values with curl before you change the editor settings.

Keep the credentials apart

An editor setup can use three credentials:

Credential	Held by	Sent to
Gateway runtime token (VERDICTAN_API_TOKEN)	The verdictan gateway run process	The Verdictan control plane
Client API token	The model client or the MCP client of the editor	The Verdictan gateway, as a Bearer token
Provider credential (secret_key_ref)	The provider target of the gateway	The upstream model provider

Do not put the gateway runtime token into an editor.

Verify that the editor uses the gateway

Send one request from the editor. Then run this command in a CLI session for the same organization and region:

verdictan events tail --since 10m --json

A matching event proves that the request went through Verdictan. If no event shows, examine these items in this sequence:

1. The gateway mode delivers events.
2. The CLI queries the same organization and region as the gateway.
3. The time window and the filters include the request.
4. The active model base URL of the editor points to the gateway.

Traffic that an editor sends directly to its vendor does not go through Verdictan.

Next steps

• Gateway Setup for IDEs: start the gateway and test the editor values.
• API Tokens for IDEs: create a scoped client token for each editor or team.
• Connect an MCP Client: the MCP endpoint, the token, and the available tools.