Skip to main content

MCP (Model Context Protocol)

SynapseKit supports the Model Context Protocol for connecting to external tool servers, wrapping MCP tools for use with agents, and exposing your own tools as an MCP server.

MCPClient​

Connect to an MCP server to discover and call remote tools.

Stdio transport​

from synapsekit import MCPClient

client = MCPClient()
await client.connect_stdio("npx", ["-y", "@modelcontextprotocol/server-filesystem", "/tmp"])

# List available tools
tools = await client.list_tools()
for t in tools:
    print(t.name, t.description)

# Call a tool
result = await client.call_tool("read_file", {"path": "/tmp/hello.txt"})
print(result)

SSE transport​

client = MCPClient()
await client.connect_sse("http://localhost:8080/sse")

tools = await client.list_tools()
result = await client.call_tool("search", {"query": "hello"})

MCPToolAdapter​

Wrap MCP tools as BaseTool instances so they work with any SynapseKit agent:

from synapsekit import MCPClient, MCPToolAdapter, FunctionCallingAgent

client = MCPClient()
await client.connect_stdio("npx", ["-y", "@modelcontextprotocol/server-filesystem", "/tmp"])

# Wrap all MCP tools as BaseTool instances
tools = await MCPToolAdapter.from_client(client)

# Use with any agent
agent = FunctionCallingAgent(llm=llm, tools=tools)
result = await agent.run("List files in /tmp")

Each wrapped tool preserves the original name, description, and parameter schema from the MCP server. The adapter handles serialization and deserialization automatically.

MCPServer​

Expose your SynapseKit tools as an MCP-compatible server so external clients can discover and call them:

from synapsekit import MCPServer, CalculatorTool, WebSearchTool

server = MCPServer(
    name="my-tools",
    tools=[CalculatorTool(), WebSearchTool()],
)

# Run as stdio server (for CLI-based MCP clients)
await server.run_stdio()

Any BaseTool subclass or @tool-decorated function can be exposed via MCP. The server automatically generates the MCP-compatible tool schema from your tool's parameters definition.

Using with Claude Desktop​

Add to your Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "synapsekit-tools": {
      "command": "python",
      "args": ["my_server.py"]
    }
  }
}

Where my_server.py contains:

import asyncio
from synapsekit import MCPServer, CalculatorTool, WebSearchTool

server = MCPServer(name="synapsekit-tools", tools=[CalculatorTool(), WebSearchTool()])
asyncio.run(server.run_stdio())