Augureaugure

MCP Server

Expose Augure's tools, memory, and personas to Claude Desktop, Cursor, and other MCP clients

Augure includes a built-in Model Context Protocol (MCP) server. This allows MCP-compatible clients like Claude Desktop, Cursor, and others to use Augure's tools, read agent memory, browse scheduled jobs, and load persona prompts.

Transports

Augure supports two MCP transports:

TransportCommandUse case
stdioaugure mcpClaude Desktop, Cursor, and other local MCP clients
HTTPaugure start --mcpRemote access, multi-client setups

The stdio transport is the standard way to connect MCP clients. The client spawns the augure mcp process and communicates over stdin/stdout.

Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "augure": {
      "command": "augure",
      "args": ["mcp", "-c", "/path/to/augure.json5"]
    }
  }
}

Cursor

Add to your Cursor MCP settings:

{
  "mcpServers": {
    "augure": {
      "command": "augure",
      "args": ["mcp", "-c", "/path/to/augure.json5"]
    }
  }
}

HTTP Transport

When running augure start --mcp or setting mcp.enabled: true in the config, Augure starts an HTTP MCP server alongside the agent:

augure start --mcp                    # default port 3100

The HTTP endpoint is available at http://localhost:3100/mcp.

Configuration

{
  mcp: {
    enabled: true,
    port: 3100,    // default
  },
}

See Configuration for the full config reference.

Capabilities

Tools

All registered Augure tools are exposed as MCP tools. The tool list depends on your configuration:

ToolDescription
memory_readRead a memory file
memory_writeWrite to a memory file
datetimeGet current date and time
scheduleCreate or manage scheduled jobs
web_searchSearch the web (requires config)
httpMake HTTP requests
emailRead and send emails (requires config)
sandbox_execExecute code in Docker container
opencodeRun a coding agent in sandbox
githubGitHub API operations (requires config)
browserBrowser automation (requires tools.browser config)
Skill toolscreate_skill, list_skills, run_skill, manage_skill, install_skill (when skills are configured)

Resources

MCP resources provide read access to agent data:

URI patternDescription
memory://<path>Memory files (markdown)
jobs://listAll scheduled jobs (JSON)
jobs://<id>Single job details (JSON)

Prompts

When personas are configured, each persona is exposed as an MCP prompt. Clients can list available personas and load them with an optional message argument for context-aware persona selection.

On this page