Connections (MCP)

Agents connect to outside services — databases, ticketing systems, internal APIs, SaaS tools — through the Model Context Protocol, the open standard for giving AI tools access to other software. If a service has an MCP server, your agents can use it.

Adding a connection

Connections are configured in ~/.openacme/mcp.json (web UI management lives under Settings). Two kinds of servers:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_TOKEN": "ghp_..." }
    },
    "internal-api": {
      "url": "https://mcp.example.com/sse",
      "headers": { "Authorization": "Bearer ..." }
    }
  }
}

• Command servers run locally as a subprocess.
• URL servers connect to a remote endpoint over HTTP/SSE.

Once connected, every tool the server offers shows up in your agents' toolboxes automatically.

Workforce-wide vs. private

Servers in mcp.json are available to the whole workforce. An agent can also have private connections — defined on the agent itself — for services only that role should touch. Your support agent's helpdesk connection doesn't need to be visible to your researcher.

An agent can likewise opt out of global connections, so access is shaped per role in both directions.

Credentials

Pass tokens explicitly via the server's env or headers — OpenAcme deliberately does not forward your shell's environment variables to MCP servers, so credentials never leak into a connection by accident.

Checking on connections

openacme mcp list      # what's configured
openacme mcp status    # what's connected right now
openacme mcp test      # try connecting and report errors