skip to content
Cursor

Your tools

Cursor

Run Cursor's chat on Conifer models, or give the agent Conifer as a set of tools. Either takes about a minute, and both can be on at once.

Cursor takes the same two facts as every other client: the base URL and your key. In a shell they are the two OpenAI variables. In Cursor they are two fields in the settings pane.

terminal
export OPENAI_BASE_URL=https://api.conifer.build/v1
export OPENAI_API_KEY=$CONIFER_API_KEY
base URL
https://api.conifer.build/v1

Run chat on Conifer models

Open Cursor Settings → Models and scroll to API Keys. Cursor documents the pane at docs.cursor.com/settings/api-keys.

  1. Paste your Conifer key into the OpenAI API Key field.
  2. Turn on Override OpenAI Base URL and enter https://api.conifer.build/v1.
  3. Click Verify. Cursor checks the key against that URL.
  4. Under the model list choose Add model and type an id from the catalog, for example claude-haiku-4-5.
  5. Pick that model in the chat’s model picker. Turning off the built-in models you are not using keeps the picker short.

That is the whole setup. Chat and agent turns on that model now run through Conifer and bill to your account, on the same key you use everywhere else.

Add Conifer as agent tools

This leaves Cursor’s models in place and gives the agent extra tools. It can ask any Conifer model mid-task, run one prompt across several models, and check what a call cost. Create .cursor/mcp.json in the project, or ~/.cursor/mcp.json for every project.

.cursor/mcp.json
{
  "mcpServers": {
    "conifer": {
      "command": "npx",
      "args": ["-y", "conifer-sdk", "conifer-mcp"],
      "env": { "CONIFER_API_KEY": "sk-conifer-…" }
    }
  }
}

Turn the server on in Cursor Settings under MCP, then ask the agent something like “run this prompt through conifer_compare on claude-haiku-4-5 and gpt-oss-20b”. The server runs from the published conifer-sdk package through npx, so there is nothing to build. The MCP server section lists each tool.

Which one to use

Use the key override when you want Cursor’s chat to run on a Conifer model, with one bill and a receipt on every turn. Use MCP when you want Cursor to keep its own models and also reach the catalog, say for a cheap model on bulk edits or a comparison before you commit to one id. The two stack.

The desktop app

Cursor is a desktop app, so the steps above are the desktop steps. The same settings pane exists on macOS, Windows and Linux, and the note about Tab autocomplete holds wherever you set the base URL. The setting does not configure the separate Cursor CLI.