Version: 1.0.0 Last Updated: 2026-05-14 Status: Active

MCP Server Documentation

This directory contains documentation for all 3 MCP (Model Context Protocol) servers in src/mcp/.

Overview

MCP servers provide standardized access to Bast AI services through the Model Context Protocol, enabling integration with AI assistants and agent frameworks.

MCP Servers

Server	Version	Description	Dependencies
base	0.1.1	Abstract base for MCP servers	fastmcp, openai-agents
mongo	0.1.1	MongoDB operations via MCP	bast-ai-mongo
ontology	0.1.1	Ontology/knowledge graph access	bast-ai-graphrag

Architecture

┌─────────────────────────────────────────┐
│              MCP Clients                │
│  (Claude, OpenAI Agents, Custom)        │
└─────────────────────────────────────────┘
                    │
                    ▼
┌─────────────────────────────────────────┐
│           bast-ai-mcp-base              │
│      (FastMCP, Common Utilities)        │
└─────────────────────────────────────────┘
         │                    │
         ▼                    ▼
┌─────────────────┐  ┌─────────────────┐
│ bast-ai-mcp-    │  │ bast-ai-mcp-    │
│     mongo       │  │    ontology     │
└─────────────────┘  └─────────────────┘
         │                    │
         ▼                    ▼
┌─────────────────┐  ┌─────────────────┐
│  bast-ai-mongo  │  │ bast-ai-graphrag│
└─────────────────┘  └─────────────────┘

Base MCP Server

The bast-ai-mcp-base provides:

• FastMCP integration (v2.10+)
• Common MCP patterns
• OpenAI agents support for testing
• Configuration management integration

Dependencies

Package	Version	Purpose
bast-ai-config	-	Configuration
fastmcp	^2.10	MCP framework
openai-agents	0.2.2	Agent testing

MCP MongoDB Server

Provides MongoDB operations through MCP:

Available Tools

Tool	Description
find_documents	Query documents with filters
insert_document	Insert new document
update_document	Update existing document
delete_document	Delete document
aggregate	Run aggregation pipeline
count_documents	Count matching documents

Usage Example

# MCP client calling MongoDB server
result = await mcp_client.call_tool(
    "find_documents",
    {
        "collection": "users",
        "filter": {"status": "active"},
        "limit": 10
    }
)

MCP Ontology Server

Provides knowledge graph access through MCP:

Available Tools

Tool	Description
query_entities	Search for entities
get_relationships	Get entity relationships
traverse_graph	Navigate knowledge graph
search_concepts	Semantic concept search

Usage Example

# MCP client calling ontology server
result = await mcp_client.call_tool(
    "query_entities",
    {
        "query": "machine learning",
        "entity_types": ["concept", "topic"],
        "limit": 5
    }
)

Running MCP Servers

Development

# Run MongoDB MCP server
pants run src/mcp/mongo:main

# Run Ontology MCP server
pants run src/mcp/ontology:main

With Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "bast-mongo": {
      "command": "python",
      "args": ["-m", "bast_ai_mcp_mongo.main"],
      "env": {
        "MONGO_URI": "mongodb://localhost:27017"
      }
    }
  }
}

Testing

Integration Tests

Each MCP server has integration tests in integration/:

• client.py - Client-side tests
• mcp_server.py - Server-side tests

# Run integration tests
pants run src/mcp/mongo/integration/client.py
pants run src/mcp/ontology/integration/mcp_server.py

Issues & Improvements

Current Issues

1. Limited tool coverage: Not all MongoDB/GraphRAG features exposed
2. No authentication: MCP servers lack built-in auth
3. Missing streaming: Large results aren't streamed

Suggested Improvements

1. Expand tool coverage: Add more MongoDB and ontology operations
2. Add authentication: Implement API key or token auth
3. Response streaming: Stream large result sets
4. Rate limiting: Add per-client rate limits
5. Caching: Add result caching for repeated queries
6. Metrics: Add Prometheus metrics endpoint

Maintainer

• Chiranjeevi G (chiranjeevi.g@cygnenoircyber.com) - All MCP servers