mcp-agent-tool-adapter

The mcp-agent-tool-adapter exposes MCP-enabled tooling that lets agents dynamically invoke modular tools. This server acts as a bridge between MCP clients and tool servers (for example, filesystem tooling) by spawning tool processes and exposing them through the MCP protocol. It supports two agent backends: Google ADK-based agents with streaming interfaces and LangGraph-based agents using ReAct-style planning with streaming tool calls. You can switch between agent types by configuring the tool_type in your client adapter. To start using it locally, ensure your Python environment is set up and point your MCP client at this server via the MCP configuration (mcp_config.json). The sample configuration demonstrates how to reference a tool package (e.g., @modular-protocol/filesystem) via npx, which the agent tool adapter will spawn and manage for you.

Prerequisites:

• Python 3.10+ installed on your system
• Git to clone the repository

Step-by-step installation:

1. Clone the repository: git clone https://github.com/serkanyasr/mcp-agent-tool-adapter cd mcp-agent-tool-adapter

2. Create a Python virtual environment (optional but recommended): python -m venv venv source venv/bin/activate # Linux/macOS venv\Scripts\activate # Windows

3. Install dependencies: pip install -r requirements.txt

4. Run the example client/server locally:

   • The repository provides Python scripts for Google ADK and LangGraph clients: python app_client_adk.py python app_client_langgraph.py

5. Prepare your mcp_config.json with the appropriate mcpServers entries (see examples in the README): { "mcpServers": { "filesystem": { "command": "npx", "args": ["@modular-protocol/filesystem"] } } }

6. (Optional) Install additional tooling or agents as needed and ensure the tool packages referenced in mcp_config.json are available in your environment.

Tips and notes:

• The server uses subprocess-based tool spawning, so ensure the referenced tool packages (e.g., @modular-protocol/filesystem) are available in your environment and can be invoked via the declared command.
• You can switch between Google ADK and LangGraph agents by adjusting the tool_type in your client configuration; both approaches are supported by the MCPClient within this adapter.
• If you encounter permission or PATH issues, ensure that node (for npx) is installed and available in your system PATH for the npx invocation to succeed.
• The example mcp_config.json demonstrates how to wire a filesystem tool; you can add more tools by duplicating the server entries under mcpServers with distinct keys.
• Review logs from the Python process for hints about tool spawning and tool adapter behavior during development.