Documentation
Cursor
Connect Milkey to Cursor with a user-wide or project-local MCP configuration.
Cursor is one of the fastest ways to use Milkey because it supports project-local and user-wide MCP configs with the same remote HTTP shape Milkey exposes.
Use a global setup when you want Milkey available across repos. Use a project-local setup when the connection should ship with one repository and stay scoped to that workspace.
Configure Cursor
The client-specific work here is mostly about where the config lives and which field names the client expects.
- Use `~/.cursor/mcp.json` for a user-wide install that follows you across projects.
- Use `.cursor/mcp.json` when the setup should live inside one repository and be reviewable by the team.
{ "mcpServers": { "milkey": { "url": "https://mcp.vexelityai.com/mcp", "headers": { "Authorization": "Bearer mk_sk_your_api_key_id_your_secret" } } }}User-wide config
Env-based header
Verify the connection
- Open Cursor MCP settings and confirm the `milkey` server appears as connected.
- Reload Cursor if you added the file manually and the server list does not update right away.
- Run a small test prompt that should require live skill retrieval rather than general model memory.
Test prompt
Use Milkey to resolve the best skill for React performance optimization in this codebase, then summarize the guidance I should follow.Operational tips
- Cursor supports variable interpolation inside `url` and `headers`, so an environment-backed token is usually the safest long-term setup.
- If a repo only occasionally needs Milkey, prefer the project-local config to avoid loading the server in unrelated workspaces.