跳到主要内容

Use SkillFlaw as an MCP server

SkillFlaw integrates with the Model Context Protocol (MCP) as both an MCP server and an MCP client.

This page describes how to use SkillFlaw as an MCP server that exposes your flows as tools that MCP clients can use when generating responses.

SkillFlaw MCP servers support both the streamable HTTP transport and Server-Sent Events (SSE) as a fallback.

For information about using SkillFlaw as an MCP client and managing MCP server connections within flows, see Use SkillFlaw as an MCP client.

Prerequisites​

Serve flows as MCP tools​

SkillFlaw exposes eligible flows through a global MCP server endpoint. Clients connect to the authenticated MCP transport at /api/v1/mcp/streamable, with /api/v1/mcp/sse available as a legacy fallback.

Flows that you want to expose as MCP tools must include a Chat Output component.

Prepare tool names and descriptions​

Tool names and descriptions help MCP clients determine which actions your flows provide and when to use those actions. It is recommended to provide clear, descriptive names and descriptions for all tools that you serve to MCP clients.

Importance of tool names and descriptions​

MCP clients use tool names and descriptions to determine which actions to use when generating responses.

Because MCP clients treat your SkillFlaw deployment as a single MCP server with available flows listed as tools, unclear names and descriptions can cause the agent to select tools incorrectly or inconsistently.

For example, a flow's default tool name is the flow ID, such as adbbf8c7-0a34-493b-90ea-5e8b42f78b66. This provides no information to an agent about the type of flow or its purpose.

To provide more context about your flows, make sure to name and describe your flows clearly before connecting clients to your SkillFlaw MCP server.

Think of these names and descriptions as function names and code comments. Use clear statements to describe the problems your flows solve.

Example: Tool name and description usage

For example, assume you create a flow based on the Document Q&A template that uses an LLM to chat about resumes, and then you give the flow the following name and description:

  • Tool name: document_qa_for_resume

  • Tool description: A flow for analyzing Emily's resume.

After connecting your SkillFlaw MCP server to Cursor, you can ask Cursor about the resume, such as What job experience does Emily have?. Using the context provided by your tool name and description, the agent can decide to use the document_qa_for_resume MCP tool to create a response about Emily's resume. If necessary, the agent asks permission to use the flow tool before generating the response.

If you ask about a different resume, such as What job experience does Alex have?, the agent can decide that document_qa_for_resume isn't relevant to this request, because the tool description specifies that the flow is for Emily's resume. In this case, the agent might use another available tool, or it can inform you that it doesn't have access to information about Alex's. For example:


_10
I notice you're asking about Alex's job experience.
_10
Based on the available tools, I can see there is a Document QA for Resume flow that's designed for analyzing resumes.
_10
However, the description mentions it's for "Emily's resume" not Alex's. I don't have access to Alex's resume or job experience information.

Connect clients to your SkillFlaw MCP server​

SkillFlaw MCP servers can be connected to local or remote MCP clients with standard MCP configuration.

The JSON option allows you to connect a SkillFlaw MCP server to any local or remote MCP client. You can modify this process for any MCP-compatible client.

  1. Install any MCP-compatible client.

    These steps use Cursor as an example, but the process is generally the same for all clients, with slight differences in client-specific details like file names.

  2. In your client, add a new MCP server using the client's UI or configuration file.

    For example, in Cursor, go to Cursor Settings, select MCP, and then click Add New Global MCP Server to open Cursor's global mcp.json configuration file.

  3. Recommended: Configure authentication for your MCP server.

  4. Add the SkillFlaw MCP server to your client's configuration file. For example:


    _11
    {
    _11
    "mcpServers": {
    _11
    "PROJECT_NAME": {
    _11
    "command": "uvx",
    _11
    "args": [
    _11
    "mcp-proxy",
    _11
    "http://SKILLFLAW_SERVER_ADDRESS/api/v1/mcp/streamable"
    _11
    ]
    _11
    }
    _11
    }
    _11
    }

    The default SkillFlaw server address is http://localhost:7860. If you are using a public SkillFlaw server, the server address is automatically included.

    If your SkillFlaw server requires authentication, you must include your SkillFlaw API key or OAuth settings in the configuration. For more information, see MCP server authentication.

  5. To include other environment variables with your MCP server command, add an env object with key-value pairs of environment variables:


    _14
    {
    _14
    "mcpServers": {
    _14
    "PROJECT_NAME": {
    _14
    "command": "uvx",
    _14
    "args": [
    _14
    "mcp-proxy",
    _14
    "http://SKILLFLAW_SERVER_ADDRESS/api/v1/mcp/streamable"
    _14
    ],
    _14
    "env": {
    _14
    "KEY": "VALUE"
    _14
    }
    _14
    }
    _14
    }
    _14
    }

  6. Save and close your client's MCP configuration file.

  7. Confirm that your SkillFlaw MCP server is on the client's list of MCP servers. If necessary, restart your client to apply the modified configuration file.

Once your MCP client is connected to your SkillFlaw MCP server, your flows are registered as tools. Cursor determines when to use tools based on your queries, and requests permissions when necessary. For more information, see the MCP documentation for your client, such as Cursor's MCP documentation.

MCP server authentication​

SkillFlaw MCP endpoints use the same authentication boundary as the SkillFlaw API.

If authentication is enabled, connect with a SkillFlaw API key or another authenticated SkillFlaw session. For MCP clients that proxy HTTP transport through mcp-proxy, include the x-api-key header in the command arguments.

For example:


_14
{
_14
"mcpServers": {
_14
"PROJECT_NAME": {
_14
"command": "uvx",
_14
"args": [
_14
"mcp-proxy",
_14
"--headers",
_14
"x-api-key",
_14
"YOUR_API_KEY",
_14
"http://SKILLFLAW_SERVER_ADDRESS/api/v1/mcp/streamable"
_14
]
_14
}
_14
}
_14
}

If SkillFlaw authentication is disabled, the MCP endpoint is publicly accessible. Only use that configuration in trusted environments.

MCP server environment variables​

MCP availability and transport behavior are controlled by your SkillFlaw deployment configuration. For the current release, use the global /api/v1/mcp/* endpoints described on this page and verify any deployment-specific environment variables against your runtime configuration.

Deploy your SkillFlaw MCP server externally​

To deploy your SkillFlaw MCP server externally, see Deploy a public SkillFlaw server.

Use MCP Inspector to test and debug flows​

Node prerequisite

MCP Inspector requires any LTS version of Node.js installed on your computer.

MCP Inspector is a common tool for testing and debugging MCP servers. You can use MCP Inspector to monitor your flows and get insights into how they are being consumed by the MCP server.

  1. Install MCP Inspector:


    _10
    npx @modelcontextprotocol/inspector

    For more information about configuring MCP Inspector, including specifying a proxy port, see the MCP Inspector GitHub project.

  2. Open a web browser and navigate to the MCP Inspector UI. The default address is http://localhost:6274.

  3. In the MCP Inspector UI, enter the connection details for your SkillFlaw MCP server. The field values depend on your server's method of authentication.

    • Transport Type: Select STDIO.
    • Command: uvx
    • Arguments: Enter the following list of arguments, separated by spaces. Replace the values for YOUR_API_KEY and SKILLFLAW_SERVER_ADDRESS with the values from your SkillFlaw deployment. For example:

      _10
      mcp-proxy --headers x-api-key YOUR_API_KEY http://SKILLFLAW_SERVER_ADDRESS/api/v1/mcp/streamable

  4. Click Connect.

    If the connection was successful, you should see the available SkillFlaw tools in the Tools tab. From this tab, you can monitor how your flows are being registered as tools by MCP, as well as test the tools with custom input values.

  5. To quit MCP Inspector, press Control+C in the same terminal window where you started it.

Troubleshoot SkillFlaw MCP servers​

For troubleshooting advice for MCP servers and clients, see Troubleshoot SkillFlaw: MCP issues.

See also​