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.
Enable the relevant Skill for an Agent and ask for the business outcome. The Agent reads the Skill, selects a tool, and invokes MCPorter from the shell. Main Agents and Background Agent Jobs use the same control-plane gateway; neither loads the entire upstream catalog as native model tools.
Configure access
For built-in GitHub, Lark, and Linear Toolkits, configure an application on the Pod Connections page. Members connect their own accounts when the operation needs personal authority. Select a named connection when several are available. Missing consent returns a link for the member to authorize and retry.
For a Skill’s remote HTTP server, configure the value named by bearer_token_env_var in Environment variables. The gateway reads that value and injects it upstream; the remote secret is excluded from Worker snapshots. Local stdio servers execute in the Worker and follow their declared command contract.
Use tools in an Applet
Publish an Action in the Applet server module. Its handler calls AnkoleRuntime.mcp.call(tool, arguments, {server, connection}); team operations use AnkoleRuntime.admin.mcp. The Function Runtime host manages invocation credentials. Use the Automation panel to create a Job, connect sources, and inspect Runs. Applet Actions do not launch Worker scripts or inherit an Agent’s enabled Skills.
Recover from failures
Use each tool’s declared input and output contract. Pass shell JSON through stdin and an explicit timeout as shown in MCP reference. Check business errors, pagination, and partial results even when transport succeeds. A partial write identifies the completed resource; repair the remaining operation instead of blindly repeating the entire request. An uncertain write outcome requires checking upstream state first.
See Automation Jobs and Environment variables.