Agent Fusion

Agent Fusion MCP tool gives multiple AI coding assistants instant access to your files—code, documents, PDFs, and more—through intelligent indexing. Self-contained deployment: runs as a single JAR with no external APIs, Docker, or cloud dependencies—just Java and a TOML config It has two independent components (each can be used alone or together):

• MCP Context Engine: RAG for Code & Documents – Local semantic search and graph traversal for AI agents. Indexes code, Word docs, PDFs, and Markdown into chunks with embeddings. Query via query_context using three parallel providers: semantic (sentence-transformers), symbol (identifier matching), full-text (keyword search). Returns ranked results with hierarchical relationships encoded as chunk IDs (parent, children, siblings, prev/next). Agents can search semantically, then traverse the knowledge graph to gather surrounding context—useful for understanding class structures, call chains, or document organization. File watcher monitors configured directories and automatically re-indexes changes, so your search index stays current as you code—no manual refresh needed. Two-file deployment (JAR + TOML config), no external dependencies, runs entirely local. Compatible with MCP protocol (works with Claudeany and any MCP-compatible AI agent). Embedding model is configurable. Includes lightweight embedding model; swap for a more powerful model if needed (see docs/other_model.md).
• Task Manager – Optionally coordinates work between multiple AIs. Routes tasks, enables voting on decisions, and tracks everything in a web dashboard.

🎥 Watch the demo to see AI assistants collaborating in action.

Why Semantic Search Matters

Most AI coding agents rely on grep and text matching. Ask them "find authentication logic" and they grep for "authentication" — missing every file that actually implements auth but calls it "login", "credentials", "token validation", or "access control".

Semantic search understands concepts, not just keywords:

• Ask for "database connection pooling" → finds connection managers, pool handlers, DB initialization
• Ask for "error handling patterns" → discovers try-catch blocks, error loggers, exception handlers
• Ask for "authentication flow" → locates login controllers, JWT validators, session managers

Even when the exact words aren't in the code.

How It Works

Agent Fusion fetches up-to-date code examples and documentation right into your LLM's context:

1️⃣ Write your prompt naturally – Ask your AI assistant what you'd normally ask

2️⃣ Tell the LLM to use query_context – Just add "use query_context to find X" in your prompt

3️⃣ Get working code answers – Instant, accurate answers based on your actual codebase

No tab-switching. No hallucinated APIs that don't exist. No outdated code generation.

Quick Start

Installation Guide – Step-by-step setup (takes 5-10 minutes)

How Search Works

The Context Engine uses three search types combined:

• Semantic Search – AI-powered understanding of meaning (finds "user authentication" when you search "login system")
• Full-Text Search – Fast keyword matching (finds exact phrases and terms)
• Hybrid Search – Combines all results ranked by relevance (you get the best matches from all methods)

Results are ranked by r

... View full README on GitHub