LLM Context Manager MCP Server & CLI wrapper using npx
npm install @senoldogann/context-manager> 🧠 The Neural Backbone for Autonomous AI Agents
Node.js wrapper for Cognitive Codebase Matrix (CCM) - Enables AI agents to understand and navigate your codebase with surgical precision.
---
1. Node.js 16+ installed
2. Ollama installed and running (for local embeddings)
- Download: https://ollama.com
- Pull required model: ollama pull mxbai-embed-large
``bash1. Install and configure for Claude Desktop, Antigravity, Cursor, etc.
npx @senoldogann/context-manager install
That's it! Restart your AI editor and start asking questions about your code.
---
📖 What is CCM?
CCM transforms static source code into a dynamic, queryable Knowledge Graph:
- 🔍 Semantic Search - Find code by meaning ("where is auth logic?")
- 🧠 Graph Navigation - Understand relationships ("who calls this function?")
- 📍 Cursor Context - Get relevant code based on your position
---
🔧 Commands
The npm wrapper downloads pre-built binaries and passes commands through:
| Command | Description |
|---------|-------------|
|
| Auto-configure MCP for editors |
| npx @senoldogann/context-manager index --path | Index a project |
| npx @senoldogann/context-manager query --text "..." | Search codebase |
| npx @senoldogann/context-manager mcp | Run MCP server directly |
| npx @senoldogann/context-manager eval --tasks | Run evaluation tasks |$3
`bash
Watch mode - auto-reindex on file changes
npx @senoldogann/context-manager index --path . --watchCustom database path
npx @senoldogann/context-manager index --path . --db-path /custom/path
`---
🔒 Privacy by Default
CCM uses a Local-First architecture:
- ✅ Your code never leaves your machine
- ✅ All embeddings run locally via Ollama
- ✅ No external API calls (unless you configure OpenAI)
---
⚙️ Configuration
$3
Create
:`ini
Local (Recommended)
EMBEDDING_PROVIDER=ollama
EMBEDDING_HOST=http://127.0.0.1:11434
EMBEDDING_MODEL=mxbai-embed-largeCloud (Optional)
EMBEDDING_PROVIDER=openai
EMBEDDING_API_KEY=sk-your-key
EMBEDDING_MODEL=text-embedding-3-smallNetworking & Limits
EMBEDDING_TIMEOUT_SECS=30
CCM_MAX_FILE_BYTES=2097152MCP Security
CCM_ALLOWED_ROOTS=/Users/you/projects:/Users/you/sandbox
CCM_REQUIRE_ALLOWED_ROOTS=0MCP Runtime
CCM_MCP_ENGINE_CACHE_SIZE=8
CCM_MCP_DEBUG=0Optional: disable embeddings entirely (semantic search disabled)
CCM_DISABLE_EMBEDDER=0Optional: embed data files (md/json/yaml) into vector search
CCM_EMBED_DATA_FILES=0Binary checksum verification (0 = enforce, 1 = bypass)
CCM_ALLOW_UNVERIFIED_BINARIES=0
`Production Tip: Set
and enable CCM_REQUIRE_ALLOWED_ROOTS=1 to restrict MCP access.---
🤖 Usage in AI
Once configured, ask your AI agent:
> "Search for the authentication flow in this codebase."
> "Read the graph for
UserService and show me its callers."> "What functions call
parse_config?"---
📦 For Developers
This package handles:
1. OS/architecture detection
2. Binary download from GitHub Releases
3. Global persistence in
~/.ccm$3
Downloads are verified against
from the GitHub Release.
If the manifest is missing or a mismatch occurs, you can set CCM_ALLOW_UNVERIFIED_BINARIES=1 to bypass verification (not recommended).$3
By default, data files (
.md, .json, .yaml) are indexed but not embedded.
Enable CCM_EMBED_DATA_FILES=1 to include them in semantic search.Source: https://github.com/senoldogann/LLM-Context-Manager
---
📝 Changelog
$3
- ✅ Release checksums (checksums.txt) for binary integrity
- ✅ MCP allowlist with optional strict enforcement
- ✅ Data file embedding is optional (CCM_EMBED_DATA_FILES`)---
Built with ❤️ in Rust