Documentation

MCP Server

Milkey’s hosted MCP path for coding agents, IDEs, and agentic developer workflows.

Milkey’s hosted MCP server gives coding agents a stable way to discover the right skill, fetch the skill body, and use that guidance at runtime without copying large prompt systems into every workspace.

Hosted endpoint

One remote MCP server URL for every supported client and IDE.

Bearer auth

One API key-based auth contract shared across every MCP client.

Core tools

Use `resolve_skill` to choose a skill and `get_skill` to fetch it.

Usage visibility

Track plan consumption and request behavior from the dashboard.

What Milkey exposes

For MCP clients, the Milkey contract is intentionally small and stable:

  • One hosted MCP endpoint at https://mcp.vexelityai.com/mcp.
  • One Bearer token authentication model backed by Milkey API keys.
  • Two core workflow tools for most agent runs: resolve_skill and get_skill.
Milkey MCP endpoint
https://mcp.vexelityai.com/mcp
When to use MCP
Use the MCP path when the primary user experience is an agent, IDE, or coding tool. If you are embedding Milkey into application code or your own provider loop, use the SDK instead.

Request flow

A typical Milkey MCP request cycle looks like this:

  1. The client connects to Milkey using your API key in the Authorization header.
  2. The agent calls resolve_skill when it needs to choose the best skill for the task.
  3. The agent calls get_skill to fetch the skill body before implementation or review work begins.
  4. Milkey records usage against your account so the dashboard can show request volume, limits, and reset timing.

Choose a client

The setup differences between clients are mostly about config shape and file location. The Milkey endpoint and auth model stay the same.

Setup guide

Cursor

Connect Milkey to Cursor with a user-wide or project-local MCP configuration.

Setup guide

Codex

Configure Milkey in Codex using the native MCP server block in your Codex config.

Setup guide

Claude Code

Add Milkey to Claude Code through the built-in `claude mcp add` workflow.

Setup guide

Windsurf

Connect Milkey to Windsurf through the MCP UI or the raw `mcp_config.json` file.

Setup guide

GitHub Copilot

Configure Milkey for GitHub Copilot in IDE-based Agent mode or Copilot CLI.

Fallback guide

All Other MCP Clients

Use the generic remote HTTP server shape when your client is not documented individually.

Authorization placement

Detailed authorization guidance should live in a dedicated MCP Authorization page rather than being repeated in full on every agent setup page.

  • Each client page should still show the real header inside the config example so setup is copyable without context switching.
  • The full explanation of key storage, rotation, malformed headers, and `401` / `400` behavior belongs in one canonical place.
  • Quickstart and client pages should link back to Authorization and API Keys instead of restating the entire auth model over and over.
Recommended structure
Keep Authorization inside the new `MCP` section. It is specific to the MCP connection path, but it should be treated as shared infrastructure for every agent page.

Shared infrastructure

Authorization

Read the shared Bearer token rules, error behavior, and storage guidance.

Key management

API Keys

Create, rotate, and revoke the keys used by Milkey MCP clients.

© 2026 Milkey MCP

SupportPrivacy