This server connects Claude to your PmWiki installation through a Docker container that reads your wiki.d directory. It exposes three tools: search_wiki for full-text queries with optional case sensitivity, read_page to pull complete page content, and list_pages to browse by group. All wiki pages show up as resources with pmwiki:// URIs. The setup runs over SSE transport and mounts your wiki directory read-only, so Claude can search and read your wiki pages without write access. You'd reach for this if you're running PmWiki and want Claude to answer questions by searching your documentation or retrieve specific page content on demand.
MCP (Model Context Protocol) server for interfacing an LLM with PmWiki.
Version française / French version
The server uses the MCP protocol with SSE (Server-Sent Events) transport to enable an LLM to interact with your PmWiki instance.
GET /sse: SSE connection to establish bidirectional communicationPOST /messages/: Receive messages from the MCP clientThe simplest method is to use the published image from Docker Hub:
docker compose up -d
If you want to build the image locally:
docker-compose.yml and comment out the image: line, then uncomment the build: . linedocker compose up -d --build
Without docker-compose, you can run directly:
docker run -d \
--name pmwiki-mcp-server \
-p 3000:3000 \
-v /path/to/your/wiki.d:/wiki_data:ro \
-e WIKI_DIR=/wiki_data \
kcofoni/pmwiki-mcp:latest
The server will be accessible at http://localhost:3000 (or http://vmtest:3000 from other machines on the network).
# Check that the server is running
docker logs pmwiki-mcp-server
# Test the SSE connection
curl -N http://localhost:3000/sse
Add this configuration to your Claude Desktop configuration file:
On macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
On Windows: %APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"pmwiki": {
"command": "mcp-proxy",
"args": [
"--transport=sse",
"http://vmtest:3000/sse"
]
}
}
}
Important notes:
vmtest is the hostname of the machine hosting the MCP server Docker containervmtest with:
localhost if Claude Desktop runs on the same machine as the server192.168.1.100:3000/sse)mcp-proxy tool must be installed (usually provided with Claude Desktop)Once connected, your LLM will have access to:
pmwiki://Group.PageNamesearch_wiki: Search for text across all pages
query (required): Text to search forcase_sensitive (optional): Case-sensitive search (default: false)read_page: Read the complete content of a page
page_name (required): Page name (e.g., Main.HomePage or Main/HomePage)list_pages: List all wiki pages
group (optional): Filter by groupThe PmWiki directory is mounted from the host machine to the Docker container:
volumes:
- /home/docker/appdata/html/wiki.d:/wiki_data:ro
Important:
/home/docker/appdata/html/wiki.d is the path on the host machine - this is an example to adaptwiki.d directory:ro) for security reasons/wiki_data is the internal container path (do not modify)The image is publicly available on Docker Hub:
kcofoni/pmwiki-mcp:latestkcofoni/pmwiki-mcp:v1.0.2To pull the latest version:
docker pull kcofoni/pmwiki-mcp:latest
To pull a specific version:
docker pull kcofoni/pmwiki-mcp:v1.0.2
wiki.d/)# View logs in real-time
docker logs -f pmwiki-mcp-server
# Last 50 lines
docker logs --tail 50 pmwiki-mcp-server
Check that the wiki directory exists:
ls -la /home/docker/appdata/html/wiki.d
Use the list_pages tool to see all available pages. The exact PmWiki filename format must be used (e.g., Main.HomePage not Main/HomePage for the file).
Check that port 3000 is properly exposed:
docker ps | grep pmwiki-mcp-server
pmwiki-mcp/
├── pmwiki_mcp_server.py # MCP server
├── requirements.txt # Python dependencies
├── Dockerfile # Docker image
├── docker-compose.yml # Docker Compose configuration
├── README.md # This documentation (English)
└── README_fr.md # French documentation
After modifying the code:
docker compose up -d --build
This project is licensed under the MIT License - see the LICENSE file for details.
WIKI_DIR*Path to the PmWiki wiki.d directory containing page files (must be mounted as read-only volume in Docker)