Fluree MCP Server (TypeScript)

A TypeScript/Node.js implementation of the Model Context Protocol (MCP) server for Fluree database access.

Features

• SPARQL Query Execution: Execute SPARQL queries with automatic FROM clause injection
• Data Agent Support: Query multiple datasets through Fluree data agents
• Data Model Caching: Efficient caching of schema information with TTL and LRU eviction
• Flexible Data Model Source: Optional separate dataset for data model vs SPARQL queries when model is not in main data dataset
• Session Management: Support for Fluree API keys and Nexus session tokens
• Error Handling: Intelligent error classification for LLM learning
• Multiple Prompts: Built-in prompts for query generation, error handling, and schema summarization

Installation

From NPM

npm install @fluree/fluree-mcp-server
# or
yarn add @fluree/fluree-mcp-server

From Source

yarn install
yarn build

Usage

Command Line

# Development mode
yarn dev

# Production mode
yarn start

# Watch mode for development
yarn watch

# CLI Arguments
yarn dev --help                                    # Show all available options
yarn dev --dataset "user/dataset"                  # Single dataset mode
yarn dev --data-agent "agent-name"                 # Data agent mode (multiple datasets)
yarn dev --data-model-dataset "company/schema"     # Separate dataset for data model
yarn dev --fluree-api-key "handle=your_api_key"    # API key authentication
yarn dev --nexus-session-token "your_token"        # Session token authentication
yarn dev --transport stdio                         # Override transport type
yarn dev --port 8080                               # Custom port
yarn dev --dataset-policy "ds1=policy1,policy2"   # Dataset policies

# Note: --dataset and --data-agent are mutually exclusive

# Environment Variables
LOG_LEVEL=debug yarn dev                           # Enable debug logging
MCP_TRANSPORT=sse yarn dev                         # Set transport via environment

Claude Desktop Integration

Add the following to your Claude Desktop configuration:

{
  "mcpServers": {
    "fluree": {
      "command": "npx",
      "args": ["@fluree/fluree-mcp-server"],
      "env": {
        "FLUREE_BASE_URL": "https://app.949824632714.flur.ee",
        "FLUREE_DATASET": "your/dataset",
        "FLUREE_API_KEY": "your-api-key",
        "FLUREE_API_HANDLE": "your-handle"
      }
    }
  }
}

Or if you prefer to install globally:

npm install -g @fluree/fluree-mcp-server

Then use:

{
  "mcpServers": {
    "fluree": {
      "command": "fluree-mcp-server",
      "env": {
        "FLUREE_BASE_URL": "https://app.949824632714.flur.ee",
        "FLUREE_DATASET": "your/dataset",
        "FLUREE_API_KEY": "your-api-key",
        "FLUREE_API_HANDLE": "your-handle"
      }
    }
  }
}

Configuration

Dataset vs Data Agent Mode

The server can operate in one of two modes:

1. Dataset Mode - Query a single specific dataset
   • Use --dataset CLI arg or FLUREE_DATASET env var
   • Example: --dataset "user/my-dataset"
2. Data Agent Mode - Query multiple datasets through a data agent
   • Use --data-agent CLI arg or FLUREE_DATA_AGENT env var
   • Example: --data-agent "my-agent"
   • The data agent handles dataset selection internally

Important: These modes are mutually exclusive. You cannot specify both a dataset and a data agent.

Environment Variables

The server supports configuration via environment variables or a .env file:

Core Configuration:

• MCP_TRANSPORT: Transport type (stdio, sse, http) - default: sse
• FLUREE_BASE_URL: Fluree instance URL
• BIND_ADDRESS: Server bind address - default: 127.0.0.1
• SERVER_PORT: Server port - default: 3456
• LOG_LEVEL: Log level (error, warn, info, debug) - default: info

Session Configuration:

• FLUREE_DATASET: Single dataset to use (conflicts with FLUREE_DATA_AGENT)
• FLUREE_DATA_AGENT: Data agent to use for accessing multiple datasets (conflicts with FLUREE_DATASET)
• DATA_MODEL_DATASET: Optional dataset to use for fetching the data model (works with either FLUREE_DATASET or FLUREE_DATA_AGENT)
• FLUREE_API_KEY: Default Fluree API key
• FLUREE_API_HANDLE: Default Fluree API handle
• NEXUS_SESSION_TOKEN: Nexus session token (alternative to API key)
• POLICY_GROUPS: JSON string of dataset policies

Legacy (Deprecated):

• STARTUP_DATASETS: Legacy comma-separated datasets (deprecated - use FLUREE_DATASET instead)

Performance Configuration:

• RATE_LIMIT_PER_MINUTE: Rate limiting - default: 100
• CIRCUIT_BREAKER_FAILURE_THRESHOLD: Circuit breaker failure threshold - default: 5
• CIRCUIT_BREAKER_TIMEOUT_SECS: Circuit breaker timeout - default: 60
• CIRCUIT_BREAKER_HALF_OPEN_MAX_CALLS: Half-open max calls - default: 3
• HTTP_REQUEST_TIMEOUT_MS: HTTP request timeout - default: 30000
• HTTP_CONNECTION_TIMEOUT_SECS: HTTP connection timeout - default: 10

Fallback Configuration:

• DEFAULT_FLUREE_API_KEY: Fallback API key
• DEFAULT_FLUREE_HANDLE: Fallback API handle

Using .env Files

Create a .env file in the project root (copy from .env.example):

# Copy the example file
cp .env.example .env

# Edit with your configuration
nano .env

Example .env file:

MCP_TRANSPORT=stdio
FLUREE_BASE_URL=https://app.949824632714.flur.ee
FLUREE_QUERY_URL=https://api.fluree.cloud
STARTUP_DATASETS=your/dataset
DATA_MODEL_DATASET=company/shared-schema
FLUREE_API_KEY=your-api-key
FLUREE_API_HANDLE=your-handle

Available Tools

get_data_model

Retrieves and caches the data model schema from configured Fluree datasets.

sparql_query

Executes SPARQL queries against session datasets with automatic FROM clause injection.

Parameters:

• query (required): The SPARQL query to execute

Available Resources

fluree://session

Current session information including datasets, policies, and cache status.

fluree://data-model

The currently cached data model schema if available.

Available Prompts

sparql_query_helper

Generate SPARQL queries from natural language descriptions using the data model.

data_model_summarizer

Create user-friendly summaries of complex data models.

sparql_error_handler

Analyze SPARQL errors and provide corrected queries.

Data Model Configuration

The server supports flexible data model sourcing:

Default Behavior

By default, the data model is fetched from the same datasets used for SPARQL queries.

Separate Data Model Dataset

You can optionally specify a different dataset for fetching the data model:

# CLI
yarn dev --dataset "company/customer-data" --data-model-dataset "company/shared-schema"

# Environment Variable
STARTUP_DATASETS=company/customer-data
DATA_MODEL_DATASET=company/shared-schema

This is useful for:

• Shared Schemas: Use a common schema across multiple data datasets
• Development: Reference production schema while querying test data
• Performance: Cache stable schema from a dedicated source

Architecture

The server is built with:

• @modelcontextprotocol/sdk: Official MCP SDK for TypeScript
• TypeScript: Type-safe development
• Node.js: Runtime environment
• Fetch API: HTTP client for Fluree API calls

Error Handling

The server distinguishes between:

• Business Logic Errors (400, 404): Returned as successful tool calls with structured error information
• Infrastructure Errors (5xx, network): Returned as tool execution failures

This approach helps LLMs learn from query mistakes while properly handling system failures.

Development

# Install dependencies
yarn install

# Start development server
yarn dev

# Build for production
yarn build

# Run built version
yarn start

License

MIT