Wayfinder-Foundry/pdal-mcp

pdal-mcp

PDAL MCP is an open-source server that exposes common PDAL command-line utilities through the Model Context Protocol (MCP). PDAL is a C++ library for reading and processing point-cloud data such as LiDAR, analogous to GDAL for raster and vector data【213350811818443†L146-L162】. MCP standardises how AI agents call external tools via JSON‑RPC, emphasising secure, discoverable and context-preserving operations【922541277255947†L51-L66】【922541277255947†L132-L152】. This project wraps PDAL’s CLI commands as MCP tools so AI assistants can inspect, translate and process point‑cloud data in a controlled environment.

Features

• Provides MCP tool definitions for key PDAL commands: info, translate, pipeline, merge, split, tile and tindex.
• Validates inputs via JSON schema and prevents access outside of a configured working directory.
• Produces structured outputs (JSON or file resources) for easy consumption by AI agents.
• Configurable through a JSON file and environment variable.
• Designed to be open-source and extensible—contributors can add more PDAL tools.

For architectural details and JSON schemas for each tool, see .

Installation

Prerequisites

• Python 3.9+.
• PDAL installed on the host system (ensure the pdal CLI is available in your PATH). You can install PDAL via conda:

conda install -c conda-forge pdal

• Clone this repository:

git clone https://github.com/JordanGunn/pdal-mcp.git
cd pdal-mcp

Install dependencies

• Recommended: use uv to manage a local Python environment and install the MCP Python SDK and PDAL bindings:

uv init  # initialise a uv project (skip if already in a uv-managed directory)
uv add "mcp[cli]" pdal

• Alternatively, install via pip:

python3 -m pip install "mcp[cli]" pdal

Running the server

This project uses the MCP Python SDK’s FastMCP server. After creating a configuration file (see below), you can run the server with uv:

# Copy the example config and edit as needed
cp config.example.json config.json

# Point the environment variable to your config
export PDAL_MCP_CONFIG=$(pwd)/config.json

# Start the MCP server in development mode with hot‑reloading
uv run mcp dev src/server.py

# Start the MCP server via stdio (useful for production)
uv run mcp src/server.py stdio

Note: Some MCP servers are distributed via Node (npx) or Rust (uvx), but this project uses the official Python SDK and uv for dependency management and execution.

Configuration

The server reads configuration from a JSON file. Set the path to the file in the PDAL_MCP_CONFIG environment variable before starting the server. A typical configuration looks like:

{
  "workdir": "/srv/pdal_mcp/workdir",
  "allowedCommands": ["info","translate","pipeline","merge","split","tile","tindex"],
  "maxTimeout": 60000,
  "port": 3001
}

• workdir – Directory where input and output files are stored. The server refuses paths outside this folder.
• allowedCommands – List of PDAL commands enabled on the server.
• maxTimeout – Maximum run time (ms) for any command.
• port – Port to bind when launching the server via uv.

MCP tools

See the design document for each tool’s input and output JSON schema.

Example usage

After starting the server on port 3001, you can call MCP methods via JSON‑RPC:

List available tools:

curl -X POST http://localhost:3001/mcp/tools/list \
 -H "Content-Type: application/json" \
 -d '{"jsonrpc":"2.0","method":"tools/list","id":1}'

Call pdal_info on a file in the working directory:

curl -X POST http://localhost:3001/mcp/calls \
  -H "Content-Type: application/json" \
  -d @- <<'EOF'
{
  "jsonrpc": "2.0",
  "method": "pdal_info",
  "params": {
    "inputFile": "data.las",
    "all": true
  },
  "id": 1
}
EOF

Contributing

CContributions are welcome! Please see for guidelines on how to report issues, propose new features and add new PDAL tools to this project.tools to this project.

Code of Conduct

This project adheres to a . By participating, you are expected to uphold this code.

License

This project is licensed under the .

Acknowledgments

• PDAL and its developers.
• Model Context Protocol community for defining the MCP specification.