Skip to content
Ankole

Use an MCP-backed Skill

For AI Agents: the Markdown version of this page is at https://ankole.agentbull.com/en-US/docs/using-mcp/index.md. The documentation index is at https://ankole.agentbull.com/en-US/llms.txt.

Ankole uses MCP behind Skills. The Skill owns domain routing and result rules. The pinned mcporter CLI owns protocol discovery and calls for the one tool that the Skill selects.

For an Agent-specific domain integration, do not register an MCP server directly on the Agent. Enable the Skill that declares it. Disabling that Skill removes the dependency from the next execution.

Main Agent and Background Agent Job

Ask the Agent for the business result, not for an MCP tool name. The Agent reads the matching Skill, chooses one tool, and inspects only that tool’s current schema when needed. It then calls mcporter with a JSON argument object on stdin.

The Main Agent uses its command tool. A Background Agent Job uses its Codex terminal. Neither path exposes the complete MCP catalog as native model tools.

Automation Job

An Automation Job does not read Skill instructions. The Agent that writes main.ts must encode the selected tool, arguments, bounds, and result checks in the script.

Each Automation attempt receives the current enabled Skill dependencies through MCPORTER_CONFIG, plus the latest Agent WorkerEnv. Call mcporter with Bun.spawn, write JSON to stdin, check the exit code, and parse stdout. Do not create ~/.mcporter/mcporter.json.

Credentials

The Skill stores a credential variable name such as MCP_HTTP_TOKEN. Configure its value in Environment variables. The generated config contains the variable name, not its value.

If a variable is missing, the call fails. Do not paste a token into chat, a Skill, a script, an argument file, or a shell command.

Failures and result bounds

An invalid declaration or conflicting server definition fails before the model command or Automation script starts. Transport, protocol, argument, and server errors produce a nonzero mcporter exit.

Command and Automation logs are bounded. Follow the Skill’s pagination, freshness, warning, and partial-result rules. A successful process exit does not prove that a business result is complete.

References