An MCP server that provides AI assistants with Ethereum network analytics capabilities via Xatu data.
Agents execute Python code in sandboxed containers with direct access to ClickHouse blockchain data, Prometheus metrics, Loki logs, and S3-compatible storage for outputs.
Read more: https://www.anthropic.com/engineering/code-execution-with-mcp
# Configure
cp config.example.yaml config.yaml
# Edit config.yaml with your datasource credentials (ClickHouse, Prometheus, Loki)
# Run (builds sandbox image, starts MinIO + MCP server)
docker-compose up -dThe server runs on port 2480 (SSE transport, configurable via MCP_SERVER_PORT) with MinIO on ports 2400/2401 (configurable via MINIO_API_PORT/MINIO_CONSOLE_PORT).
See docs/deployments.md for dev, local-agent, and remote-agent deployment modes, plus separation-of-concerns notes.
Add to ~/.claude.json under mcpServers:
{
"ethpandaops-mcp": {
"type": "http",
"url": "http://localhost:2480/mcp"
}
}Install skills to give Claude knowledge about querying Ethereum data:
npx skills add ethpandaops/mcpThis installs the query skill which provides background knowledge for using the MCP tools effectively (ClickHouse queries, Prometheus metrics, Loki logs, session management, etc.).
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"ethpandaops-mcp": {
"type": "http",
"url": "http://localhost:2480/mcp"
}
}
}| Tool | Description |
|---|---|
execute_python |
Execute Python in a sandbox with the ethpandaops library |
search_examples |
Search for query examples and patterns |
Resources are available for getting started (mcp://getting-started), datasource discovery (datasources://), network info (networks://), table schemas (clickhouse://), and Python API docs (python://ethpandaops).
make build # Build binary
make test # Run tests
make lint # Run linters
make docker # Build Docker image
make docker-sandbox # Build sandbox imageMIT