MCP server

Connect codemix to AI clients that support remote MCP and need OAuth-scoped project context.

The MCP server lets an AI client ask codemix for project context directly.

Use it when copying background into prompts is the wrong abstraction. Which is often.

Endpoint

https://codemix.com/mcp

Configure it as a remote HTTP MCP server in clients that support that transport.

{
  "mcpServers": {
    "codemix": {
      "type": "http",
      "url": "https://codemix.com/mcp"
    }
  }
}

What it is for

MCP is for AI tools that need product intent while they work.

The main use case is project-scoped Q&A:

{
  "question": "What rules govern user invitations in this project?"
}

The answer should come from the codemix project model, not from the model guessing from whatever context happened to be in the chat.

Authorization

codemix MCP uses OAuth.

The setup flow asks you to choose the organization and project the client can access. That scope matters. An MCP connection should not become vague account-wide access.

Normal flow:

  1. add the codemix MCP server in your client
  2. trigger a request that needs access
  3. complete OAuth in the browser
  4. choose the organization and project
  5. return to the client

If answers refer to the wrong product, re-run authorization and check the selected project.

Capabilities

The server exposes tools and resources for project-context access.

Current scopes include:

mcp.tools.read
mcp.tools.execute
mcp.resources.read

Prompts are disabled. The useful surface is tools and resources.

Primary tool

Use ask_codemix for domain and product-context questions.

Ask about one thing at a time:

What happens when a proposal is published?
Which concepts does hosted discovery create?
What plan limits affect Slack setup?
Which screens are involved in task review?

If you want implementation instructions, ask codemix for the product rule first, then let your coding agent decide how to change the code.

MCP vs CLI

Use MCP when the AI client already supports remote tools and should authenticate through OAuth.

Use the CLI when you are working in a terminal, running local discovery, claiming tasks or scripting task execution.

Use API keys when you are building automation outside an MCP client.

Troubleshooting

Check these first:

  1. the server URL is https://codemix.com/mcp
  2. OAuth completed successfully
  3. the selected organization and project are correct
  4. the project has enough discovery context
  5. your plan allows the feature you are trying to use

If connection works but answers are weak, the problem is usually project context, not MCP transport.