OriginTrail/dkg-mcp-server

OriginTrail DKG MCP Server (example)

Overview

The OriginTrail DKG MCP Server connects MCP-compatible agents with the OriginTrail Decentralized Knowledge Graph (DKG), making it easy to create, retrieve, link, and exchange verifiable knowledge.

Note: This is BETA software and not recommended for use in production

Key Features

• SPARQL Querying: Retrieve knowledge from the DKG using flexible SPARQL queries.
• Knowledge Asset Creation: Convert natural language into structured, schema.org-compliant JSON-LD and publish it to the DKG.
• Agent Memory: Store and retrieve decentralized agent memory in a standardized, interoperable way.
• Interoperability: Works with any MCP-compatible client, including VS Code, Cursor, Microsoft Copilot agents, and more.

Getting Started

1. Clone the Repository

git clone <repo-url>
cd otdkg-mcp-server

2. Install Dependencies

Ensure you have Python 3.10+ installed. Then run:

pip install -r requirements.txt

3. Configure Environment Variables

Copy .env.example to .env and fill in the required values:

• ORIGINTRAIL_NODE_URL: You can use the default public node on testnet, use a different public testnet or mainnet node, or deploy and use your own Edge Node.
• BLOCKCHAIN: Blockchain to use for publishing Knowledge Assets on the DKG (e.g., NEUROWEB_TESTNET)
• PRIVATE_KEY: Private key of the wallet you'll use for publishing Knowledge Assets to the DKG
• GOOGLE_API_KEY: API key for Google Generative AI (you can get your API ke at https://aistudio.google.com/)

See .env.example for detailed comments and options.

4. Run the MCP Server

You can run the server in two modes:

a) Stdio Mode (for local clients like VS Code, Cursor, Claude, etc.)

python dkg_server.py --transport stdio

b) SSE Mode (for server deployment, making the DKG MCP server accessible to e.g. Microsoft Copilot Studio agents)

python dkg_server.py --transport sse

The SSE server will listen on the configured host and port (see .env).

Usage

Once the server is running, you can import it into your client and gain access to the following out-of-the-box tools:

• Query the DKG: Use the query_dkg_by_name tool to search for entities by name using SPARQL.
• Create Knowledge Assets on the DKG: Use the create_knowledge_asset tool to convert natural language into JSON-LD and publish it to the DKG.

These tools are exposed via MCP and can be invoked from any compatible agent or client.

Compatible Clients

• VS Code
• Cursor
• Claude
• Microsoft Copilot Studio agents
• Any MCP-compatible LLM or agentic framework

Extending the Server

• Customize Existing Tools: Modify and enhance the existing tools in dkg_server.py or add new functionality to tailor them to your needs.
• Add New Tools: You can easily add new MCP tools by defining new functions in dkg_server.py using the @mcp.tool() decorator (e.g. a tool that will transform website URLs into knowledge on the DKG).
• Custom Prompts: Modify or add prompt templates in the prompts/ directory to customize LLM behavior.
• Contribute: Clone, enhance, and submit pull requests to add new features or tools. Community contributions are welcome!

Project Structure

• dkg_server.py — Main server and tool definitions
• prompts/ — Prompt templates for LLM-powered tools
• requirements.txt — Python dependencies
• .env.example — Example environment configuration
• origintrail-dkg-mcp.yaml — OpenAPI spec for SSE deployment

Empower your agents to create, retrieve, and exchange verifiable knowledge on OriginTrail DKG!