@dewtech/dare-mcp-server

@dewtech/dare-mcp-server

Model Context Protocol (MCP) server for DARE Framework. Provides semantic context queries to reduce token consumption by up to 90%.

Features

• HTTP API implementing Model Context Protocol
• Semantic context queries without reading full files
• Task status management and tracking
• Blueprint and DAG access endpoints
• Full-text search with FTS5

Installation

npm install @dewtech/dare-mcp-server

Usage

Start the server

dare-mcp-server
# or
DARE_MCP_PORT=3000 DARE_PROJECT_PATH=/path/to/project dare-mcp-server

Query context (saves tokens)

# Instead of reading BLUEPRINT.md entirely, query only what you need
curl -X POST http://localhost:3000/context/query \
  -H "Content-Type: application/json" \
  -d '{"type": "architecture", "query": "authentication", "limit": 3}'

Update task status

curl -X PUT http://localhost:3000/tasks/task-001 \
  -H "Content-Type: application/json" \
  -d '{"status": "DONE"}'

Endpoints

Method	Endpoint	Description
GET	/health	Health check
GET	/tools	List MCP tools
POST	/context/query	Semantic context query
GET	/blueprint	Get BLUEPRINT.md
GET	/dag	Get dare-dag.yaml
GET	/tasks/:id	Get task status
PUT	/tasks/:id	Update task status
GET	/project	Get dare.config.json

Integration with Cursor

Add to .cursorrules:

## MCP Server
Query context via MCP instead of reading full files:
POST http://localhost:3000/context/query
{"type": "architecture", "query": "your query", "limit": 5}
This saves ~90% of tokens on context retrieval.

Token Savings

Method	Tokens Used
Read full BLUEPRINT.md	~8,000 tokens
MCP query (5 results)	~400 tokens
Savings	~95%