Register with Claude clients

Once you've built an MCP server (in Python or TypeScript), point your MCP client at it. The configuration is similar across clients but the file location and field names differ. This page shows the snippet for each.

Claude Desktop

Edit the claude_desktop_config.json file in the platform-specific location:

For the Python server:

{
  "mcpServers": {
    "axiom": {
      "command": "uv",
      "args": ["run", "/absolute/path/to/axiom_mcp.py"],
      "env": { "AXIOM_API_KEY": "your-api-key-here" }
    }
  }
}

For the Node server:

{
  "mcpServers": {
    "axiom": {
      "command": "node",
      "args": ["/absolute/path/to/axiom_mcp.cjs"],
      "env": { "AXIOM_API_KEY": "your-api-key-here" }
    }
  }
}

Restart Claude Desktop. Your tools appear in the tool list and the LLM can call them.

Claude Code

Claude Code reads MCP servers from a project-level .mcp.json file or from a global config. The simplest setup is a per-project file:

{
  "mcpServers": {
    "axiom": {
      "command": "node",
      "args": ["/absolute/path/to/axiom_mcp.cjs"],
      "env": { "AXIOM_API_KEY": "your-api-key-here" }
    }
  }
}

Place this at the root of the project and Claude Code picks it up automatically.

Cursor

Cursor uses a per-project or global mcp.json file. The schema is the same as Claude Desktop's:

{
  "mcpServers": {
    "axiom": {
      "command": "node",
      "args": ["/absolute/path/to/axiom_mcp.cjs"],
      "env": { "AXIOM_API_KEY": "your-api-key-here" }
    }
  }
}

Restart Cursor. Tools appear in the chat side panel.

Cowork

Cowork registers MCP servers via its plugin marketplace or via direct config in the desktop app. For a custom server:

1. Open Cowork settings.
2. Go to MCP servers.
3. Click Add server and paste the same config object as above.

Cowork hot-reloads the server, no restart needed.

Verify the registration worked

After adding the config and restarting, ask the LLM:

What tools do you have available?

It should list trigger_automation and run_status (or whatever names you chose). If they don't appear:

• Check the path is absolute. Relative paths break because the client runs the server from its own working directory.
• Check the env var is set. Most failures come from a missing AXIOM_API_KEY. Restart the client after adding it.
• Check the client logs. Each client surfaces server stdout/stderr somewhere, look for connection errors or missing dependencies.

Security

The MCP config file holds your API key in plain text. A few precautions:

• Don't commit the config file to a public repo.
• Use a secrets manager or a .env file the server reads, rather than hard-coding the key in the config, if your client supports it.
• Rotate your axiom.ai API key (see Authentication) if you suspect the config has been exposed.

Notes

• Each client uses the stdio transport by default for local servers. For a server running on another machine, use streamable HTTP and update the config to point at the URL instead of a command.
• Multiple mcpServers entries can coexist in a single config, your axiom.ai server can sit alongside servers for filesystem, GitHub, Slack, and so on.
• Restart the client after editing the config. Most clients don't watch the config file for changes.