mcp-server-sparql

A generic Model Context Protocol (MCP) server that connects AI agents to any SPARQL 1.1 endpoint.

Works with Apache Jena Fuseki · GraphDB · Stardog · Virtuoso · Amazon Neptune · Blazegraph — or any endpoint that speaks SPARQL 1.1.

Tools

Tool	Description
sparql_select	Execute a SELECT query → tabular rows; paginate with limit/offset/has_more
sparql_ask	Execute an ASK query → boolean; fast existence/reachability checks
sparql_construct	Execute a CONSTRUCT query → RDF graph string
sparql_describe	DESCRIBE a resource URI → all triples about it
sparql_update	Any SPARQL 1.1 Update: INSERT DATA, DELETE, CLEAR, DROP, LOAD, COPY, MOVE, ADD
graph_list	List all named graphs with triple counts
graph_get	Fetch a named graph as Turtle; paginated
graph_put	Replace a named graph (inline content or remote URL via server-side LOAD)
graph_post	Append to a named graph
graph_delete	Delete a named graph
validate_shacl	Validate RDF against SHACL shapes — pure function, no endpoint needed
endpoint_ping	Check connectivity; returns response time and configured URLs

Connection is configured once at the server level via environment variables. Tools take no connection parameters — the same pattern used by mcp-server-postgres and mcp-server-sqlite.

Quickstart

git clone https://github.com/Kashif-Rabbani/mcp-server-sparql.git
cd mcp-server-sparql
pip install -e .

cp .env.example .env
# edit .env — set SPARQL_ENDPOINT (or the three explicit URL vars)

python server.py

Server starts on http://0.0.0.0:8000. The SSE endpoint is at /sse.

Configuration

Copy .env.example to .env and set one of:

Option A — single base URL (Fuseki / Stardog):

SPARQL_ENDPOINT=http://localhost:3030/myDataset

This auto-derives /query, /update, and /data suffixes.

Option B — explicit URLs (GraphDB, Virtuoso, Neptune, etc.):

SPARQL_QUERY_URL=http://localhost:7200/repositories/myrepo
SPARQL_UPDATE_URL=http://localhost:7200/repositories/myrepo/statements
SPARQL_GRAPH_STORE_URL=http://localhost:7200/repositories/myrepo/rdf-graphs/service

Auth (optional):

SPARQL_USERNAME=admin
SPARQL_PASSWORD=secret

Endpoint-specific notes

Endpoint	Option	Query URL pattern
Apache Jena Fuseki	A	http://host:3030/<dataset>
Stardog	A	http://host:5820/<database>
GraphDB	B	http://host:7200/repositories/<repo>
Amazon Neptune	B	https://<cluster>.neptune.amazonaws.com:8182/sparql
Virtuoso	B	http://host:8890/sparql
Blazegraph	B	http://host:9999/blazegraph/namespace/<ns>/sparql

Connect to an MCP client

Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "sparql": {
      "command": "uvicorn",
      "args": ["server:app", "--host", "0.0.0.0", "--port", "8000"],
      "cwd": "/path/to/mcp-server-sparql",
      "env": {
        "SPARQL_ENDPOINT": "http://localhost:3030/myDataset"
      }
    }
  }
}

Vectara Agent Platform

Register the server as an MCP Tool Server pointing at the public SSE URL (use ngrok for local development):

ngrok http 8000
# then register https://<id>.ngrok-free.app/sse as an MCP tool server in Vectara

Requirements

• Python 3.11+
• A running SPARQL 1.1 endpoint

License

MIT