MCP Client Component

Use the MCP Client component to enable your agent to communicate with external tools, systems, and data sources that support the Model Context Protocol (MCP). This component acts as a universal adapter, allowing your agent to interact with a growing ecosystem of MCP-enabled services.

Why this matters

The Model Context Protocol standardizes how AI applications exchange information with external systems. By using the MCP Client, you can connect your agent to specialized third-party tools for data analysis, function execution, or document access through a single, unified interface, dramatically expanding its capabilities without needing custom integrations.

What You’ll Configure

• Important Limitations
• Configure the Connection
• Provide Inputs
• Handle the Output
• Best Practices
• Troubleshooting Tips
• What to Try Next

Step 1: Important Limitations

Please be aware of the current limitations of this component.

Current Limitations

• The MCP Client component currently only supports connecting to MCP servers that are publicly exposed on the web via an SSE (Server-Sent Events) URL.
• It does not support connecting to MCP servers running locally on your computer. Support for local servers is planned for a future release of the SRE Local CLI.

Step 2: Configure the Connection

Set up the details for connecting and interacting with the external MCP server.

Setting	Required?	Description
MCP URL	Yes	The Server-Sent Events (SSE) URL of the MCP server you want to integrate with.
Prompt	Yes	The natural language instruction that tells the MCP server what task to perform. You can use variables like {{prompt}} to inject dynamic content from other inputs.
Model	Yes	Select the LLM that will be used to process requests and interact with the MCP server.
System Prompt (Advanced)	No	An advanced setting that defines overarching behavior instructions for the client, similar to a system prompt in an LLM, providing context for how all requests should be handled.

Step 3: Provide Inputs

The component has a default Prompt input and allows for custom inputs to make your requests dynamic.

Input	Description
Prompt	The default input that is automatically injected into the Prompt setting using the {{prompt}} variable. Connect another component's output here to pass dynamic instructions.
Custom Inputs	Click the + icon to add new inputs (e.g., query, file_id). You can reference these in your Prompt or System Prompt settings.

Step 4: Handle the Output

The component provides a single Output containing the full response from the MCP server and allows you to define custom outputs to easily parse this response.

Output	Description
Output	The default output containing the complete, structured response from the MCP server.
Custom Outputs	Click the + icon to add new outputs that extract specific data from the main Output (e.g., summary, status). SmythOS automatically handles the parsing, so you only need to provide a name.

Step 5: Best Practices

• Verify Your MCP Server: Before integrating, ensure the external MCP server is running and its SSE URL is correct and publicly accessible.
• Use a System Prompt for Context: If you need the MCP server to behave in a specific way across all requests (e.g., "always respond in JSON"), use the System Prompt field.
• Structure Your Prompts: Write clear and direct prompts. If the MCP server supports multiple tools, specify which one you intend to use.
• Use Custom Outputs: For MCP servers that return structured data (like JSON), use custom outputs to neatly extract the specific fields you need for downstream components. This is much cleaner than parsing the raw Output with a Code component.

Step 6: Troubleshooting Tips

If your MCP Client is not responding correctly...

• Connection Fails: Double-check the MCP URL. Ensure it is a valid, publicly accessible SSE endpoint. The component cannot connect to local or private URLs.
• Incorrect Response: The MCP server's response depends heavily on your prompt. If the response is not what you expect, refine your prompt to be more specific. Test in Debug Mode to inspect the raw Output from the server.
• Authentication Issues: If the MCP server requires authentication, you may need to pass an API key or token as part of your prompt or within a custom input, depending on the server's specific implementation.

Step 7: What to Try Next

• Connect this component to a third-party MCP-enabled tool to extend your agent's capabilities with specialized functions.
• Use a GenAI LLM Component to generate a complex prompt, and then feed that prompt into the MCP Client.
• Parse the Output from the MCP server using a JSON Filter Component to clean the data before storing it or using it in another step.