Rune-stone

MCP Server Integration Example

A complete Model Context Protocol (MCP) server with tools defined by RUNE specifications.

What This Demonstrates

Files

File Description
search_documents.rune RUNE spec for the semantic search tool
file_operations.rune RUNE spec for the file reading tool
server.py Complete MCP server implementing both tools

Tools

search_documents

Searches an indexed document collection using keyword matching (swap in a vector DB for production). Supports filtering by date range, tags, and author.

results = await search_documents("python async", max_results=5)

read_file

Reads file contents with size limits, encoding support, and path traversal protection.

result = await read_file("docs/readme.md", max_bytes=4096)

Setup

Prerequisites

pip install "mcp>=1.0.0"

For production semantic search, also install:

pip install "chromadb>=0.4.0" "sentence-transformers>=2.2.0"

Run the Server

python server.py

Test with MCP Inspector

npx @modelcontextprotocol/inspector python server.py

Configure for Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "runestone-example": {
      "command": "python",
      "args": ["path/to/examples/integrations/mcp-example/server.py"],
      "env": {
        "RUNE_SANDBOX_ROOT": "/path/to/allowed/directory"
      }
    }
  }
}

Development Workflow

  1. Design – Write RUNE specs for each tool
  2. Review – Validate specs cover all edge cases and error conditions
  3. Implement – Build server.py following the specs exactly
  4. Test – Verify all TESTS from the specs pass
  5. Deploy – Register the MCP server with your AI assistant

Extending This Example

To add a new tool:

  1. Create a .rune spec file defining the tool contract
  2. Implement the async tool function in server.py following the spec
  3. Add a Tool entry in list_tools() with a JSON Schema
  4. Handle the tool name in call_tool()
  5. Test all cases from the TESTS section
This site is open source. Improve this page.