Connect MCP Tools
MCP (Model Context Protocol) is an open protocol for connecting external tools and services to an AI agent. Once you connect an MCP service, the tools it provides appear in Silicon Worker for the agent to call directly during tasks, extending what it can do.
What an MCP connector is
An MCP connector corresponds to one external service. Silicon Worker acts as an MCP client; once connected, it brings the tools that service exposes into the set of available tools. It supports these connection methods:
- stdio: launch a program on your machine and communicate over standard input/output. Good for local command-line tools.
- HTTP: connect to a service at a remote address.
- SSE: a streaming connection based on Server-Sent Events, for remote services that support it.
Add an MCP connector
- Open Silicon Worker → Settings → MCP.
- Add a connector and fill it in according to the service type:
- stdio: enter the launch command and any required environment variables.
- HTTP / SSE: enter the service address (URL) and any required headers.
- Save. Silicon Worker will try to connect and read the tools the service provides.
Follow the service's docs
The exact launch command, address, environment variables, and headers should follow the official documentation of the MCP service you're connecting to.
When authorization (OAuth) is required
Some services require sign-in authorization before access. Silicon Worker supports OAuth: when a connector needs authorization, follow the on-screen prompts to complete the sign-in flow. The credentials are stored locally for later connections.
How the tools become available
Once connected, the tools that service provides are added to the agent's available tools. You don't need to specify them manually — describe your need clearly in the task and the agent will call those tools when needed.
Risk and confirmation
Tools added via MCP are still subject to Silicon Worker's permission and risk controls. For sensitive actions, it pauses and asks for your consent first. Only connect services you trust.
Result and verification
When the connector shows a healthy status in Settings → MCP and lists the tools it provides, the connection succeeded. Then give it a task that needs that tool and watch whether the agent calls it successfully.
Related
- Understand tools and risk: Tools & risk
- Built-in capabilities: Built-in tools
- Package capabilities with plugins: Use and manage Skills
- Guides overview: Guides
