Files
syndarix/mcp-servers
Felipe Cardoso d0fc7f37ff feat(knowledge-base): implement Knowledge Base MCP Server (#57)
Implements RAG capabilities with pgvector for semantic search:

- Intelligent chunking strategies (code-aware, markdown-aware, text)
- Semantic search with vector similarity (HNSW index)
- Keyword search with PostgreSQL full-text search
- Hybrid search using Reciprocal Rank Fusion (RRF)
- Redis caching for embeddings
- Collection management (ingest, search, delete, stats)
- FastMCP tools: search_knowledge, ingest_content, delete_content,
  list_collections, get_collection_stats, update_document

Testing:
- 128 comprehensive tests covering all components
- 58% code coverage (database integration tests use mocks)
- Passes ruff linting and mypy type checking

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-03 21:33:26 +01:00
..

Syndarix MCP Servers

Model Context Protocol (MCP) servers providing tool access for Syndarix agents.

Per ADR-005, all tools require explicit project scoping.

Server Overview

Server Priority Purpose Phase
llm-gateway 1 LLM routing with failover and cost tracking Phase 2
knowledge-base 2 RAG with pgvector for semantic search Phase 2
git 3 Git operations (clone, commit, push, PR) Phase 2
issues 4 Issue tracker sync (Gitea, GitHub, GitLab) Phase 2
filesystem 5 Sandboxed file operations Phase 5
code-analysis 6 AST parsing, linting, type checking Phase 5
cicd 7 CI/CD pipeline management Phase 5

Architecture

Each MCP server is a FastMCP application that:

  1. Exposes tools via Model Context Protocol
  2. Requires project_id for all operations (explicit scoping)
  3. Uses Redis for pub/sub communication with agents
  4. Logs all operations to PostgreSQL for audit
┌─────────────────────────────────────────────────────────────┐
│                    Agent Runner                              │
│                         │                                    │
│           ┌─────────────┼─────────────┐                     │
│           ▼             ▼             ▼                     │
│    ┌──────────┐  ┌──────────┐  ┌──────────┐                │
│    │ LLM GW   │  │   Git    │  │ Issues   │  ... (7 total) │
│    │   MCP    │  │   MCP    │  │   MCP    │                 │
│    └──────────┘  └──────────┘  └──────────┘                 │
└─────────────────────────────────────────────────────────────┘

Running Locally

Each MCP server runs as a separate Docker container. See docker-compose.dev.yml for configuration.

# Start all MCP servers (Phase 2+)
docker-compose -f docker-compose.dev.yml up -d llm-gateway knowledge-base git issues

# View logs
docker-compose logs -f llm-gateway

Development

Each server follows the FastMCP pattern:

from fastmcp import FastMCP

mcp = FastMCP("server-name")

@mcp.tool()
def my_tool(project_id: str, ...):
    """Tool with required project scoping."""
    # Validate project access
    # Execute operation
    # Log for audit
    pass

See individual server READMEs for specific tool documentation.