Content
[](https://github.com/m1rl0k/Context-Engine/actions/workflows/ci.yml)
[](https://www.npmjs.com/package/@context-engine-bridge/context-engine-mcp-bridge)
[](https://marketplace.visualstudio.com/items?itemName=context-engine.context-engine-uploader)
**Documentation:** [Getting Started](docs/GETTING_STARTED.md) · README · [Configuration](docs/CONFIGURATION.md) · [IDE Clients](docs/IDE_CLIENTS.md) · [MCP API](docs/MCP_API.md) · [ctx CLI](docs/CTX_CLI.md) · [Memory Guide](docs/MEMORY_GUIDE.md) · [Architecture](docs/ARCHITECTURE.md) · [Multi-Repo](docs/MULTI_REPO_COLLECTIONS.md) · [Observability](docs/OBSERVABILITY.md) · [Kubernetes](deploy/kubernetes/README.md) · [VS Code Extension](docs/vscode-extension.md) · [Troubleshooting](docs/TROUBLESHOOTING.md) · [Development](docs/DEVELOPMENT.md)
---
## Context-Engine
Open-core, self-improving code search that gets smarter every time you use it.
---
## Quick Start
### Install the CLI
```bash
# Using pip
pip install context-engine
# Or using uv (recommended)
uv pip install context-engine
```
This installs the `ctx` command (also available as `ctx-cli`).
### One Command Setup
```bash
ctx quickstart
```
This single command starts all services, indexes your codebase, and warms up models.
### Common Commands
```bash
ctx status # Check service health
ctx reset --mcp # Reset with HTTP MCP endpoints (for Codex, modern clients)
ctx reset --mcp --db-reset # Hard reset: wipe database and rebuild everything
ctx search "auth flow" # Search your codebase
ctx answer "How does caching work?" # Get answers with citations
```
See [ctx CLI Reference](docs/CTX_CLI.md) for all commands.
<details>
<summary><b>Legacy: Makefile Commands</b></summary>
The ctx CLI deprecates the Makefile, but legacy commands retest available:
```bash
git clone https://github.com/m1rl0k/Context-Engine.git && cd Context-Engine
make bootstrap # Equivalent to: ctx quickstart
make reset # Equivalent to: ctx reset --mcp
```
</details>
---
### Enterprise-Ready Features
- **Built-in authentication** with session management (optional)
- **Unified MCP endpoint** that combines indexer and memory services
- **Automatic collection injection** for workspace-aware queries
**Direct HTTP endpoints:**
```json
{
"mcpServers": {
"qdrant-indexer": { "url": "http://localhost:8003/mcp" },
"memory": { "url": "http://localhost:8002/mcp" }
}
}
```
*See [docs/IDE_CLIENTS.md](docs/IDE_CLIENTS.md) for MCP configuration examples and [docs/MCP_API.md](docs/MCP_API.md) for the complete API reference.*
---
## Agent Skills (Codex / Claude / Gemini / Augment)
Context-Engine includes agent skills that teach AI coding assistants how to use the MCP tools effectively.
**For Claude Code:**
Install the skill plugin via marketplace:
```bash
/plugin marketplace add m1rl0k/Context-Engine
/plugin install context-engine
```
Or copy the CLAUDE.md file manually:
```bash
# Project-specific
cp CLAUDE.md /path/to/your/project/
# Or global
cp CLAUDE.md ~/.claude/CLAUDE.md
```
**For Codex (OpenAI):**
Install directly from GitHub using the `$skill-installer`:
```t
$skill-installer install https://github.com/Context-Engine-AI/Context-Engine/tree/test/.codex/skills/context-engine
```
Or copy manually:
```bash
# Global install
cp -r .codex/skills/context-engine ~/.codex/skills/
# Or project-specific
cp -r .codex/skills/context-engine /path/to/your/project/.codex/skills/
```
Restart Codex after installing to pick up the new skill.
**For Claude:**
```bash
cp CLAUDE.md /path/to/your/project/
```
**For Gemini:**
```bash
cp GEMINI.md /path/to/your/project/
```
**For Augment:**
```bash
cp -r .augment /path/to/your/project/
```
Skills teach agents to prefer Context-Engine MCP tools over grep/find/cat for code exploration.
---
## How It Works
### Local Mode Architecture
*Zero auth, single-tenant, perfect for personal use*
```mermaid
flowchart TB
subgraph client["CLIENT LAYER"]
direction LR
IDE["<b>IDE / AI Tool</b><br/>Claude, Cursor, etc."]
VSC["<b>VS Code Extension</b><br/>File Watcher"]
end
subgraph mcp["MCP LAYER"]
direction LR
subgraph search["Search Services"]
MCP_SSE["<b>Memory MCP</b><br/>:8000 SSE"]
MCP_HTTP["<b>Memory MCP</b><br/>:8002 RMCP"]
end
subgraph index["Index Services"]
IDX_SSE["<b>Indexer MCP</b><br/>:8001 SSE"]
IDX_HTTP["<b>Indexer MCP</b><br/>:8003 RMCP"]
end
end
subgraph processing["PROCESSING LAYER"]
direction LR
EMB["<b>Embedding Service</b><br/>:8100-8101<br/>ONNX (2 replicas)"]
LLAMA["<b>LLM Decoder</b><br/>:8080<br/>llama.cpp"]
LEARN["<b>Learning Worker</b><br/>Background<br/>Adaptive Reranker"]
WATCH["<b>Watcher</b><br/>File Monitor<br/>Auto-reindex"]
end
subgraph storage["STORAGE LAYER"]
direction LR
QDRANT[("<b>Qdrant</b><br/>:6333/:6334<br/>Vector DB")]
REDIS[("<b>Redis</b><br/>:6379<br/>Cache/State")]
FS[("<b>Filesystem</b><br/>/work<br/>Direct Access")]
end
%% Client connections
IDE -->|"MCP Protocol"| MCP_SSE & MCP_HTTP
IDE -->|"MCP Protocol"| IDX_SSE & IDX_HTTP
VSC -->|"File Sync"| FS
%% MCP to Processing
MCP_SSE & MCP_HTTP -->|"embed()"| EMB
IDX_SSE & IDX_HTTP -->|"embed()"| EMB
IDX_SSE & IDX_HTTP -->|"expand query"| LLAMA
%% Processing to Storage
EMB -->|"vectors"| QDRANT
LEARN -->|"update weights"| QDRANT
WATCH -->|"file events"| FS
WATCH -->|"reindex"| QDRANT
%% Storage connections
MCP_SSE & MCP_HTTP -->|"search"| QDRANT
IDX_SSE & IDX_HTTP -->|"upsert"| QDRANT
IDX_SSE & IDX_HTTP -->|"state"| REDIS
%% Styling - GitHub light/dark compatible
classDef clientStyle fill:#4a90d9,stroke:#2563eb,stroke-width:2px,color:#fff
classDef mcpStyle fill:#8b5cf6,stroke:#6d28d9,stroke-width:2px,color:#fff
classDef processStyle fill:#f59e0b,stroke:#d97706,stroke-width:2px,color:#000
classDef storageStyle fill:#10b981,stroke:#059669,stroke-width:2px,color:#fff
class IDE,VSC clientStyle
class MCP_SSE,MCP_HTTP,IDX_SSE,IDX_HTTP mcpStyle
class EMB,LLAMA,LEARN,WATCH processStyle
class QDRANT,REDIS,FS storageStyle
```
<details>
<summary><b>Service Port Reference</b></summary>
| Service | Port | Protocol | Description |
|---------|------|----------|-------------|
| Memory MCP (SSE) | 8000 | SSE | Legacy streaming transport |
| Memory MCP (RMCP) | 8002 | HTTP | Modern streamable HTTP |
| Indexer MCP (SSE) | 8001 | SSE | Legacy streaming transport |
| Indexer MCP (RMCP) | 8003 | HTTP | Modern streamable HTTP |
| Upload Service | 8004 | HTTP | Delta bundle uploads |
| Embedding Service | 8100-8101 | HTTP | ONNX embeddings (2 replicas) |
| LLM Decoder | 8080 | HTTP | llama.cpp query expansion |
| Qdrant | 6333/6334 | HTTP/gRPC | Vector database |
| Redis | 6379 | TCP | Cache and state backend |
</details>
---
## Benchmarks
### CoSQA (Dense Retrieval, No Rerank)
| Method | MRR | R@1 | R@5 | R@10 | NDCG@10 |
|--------|-----|-----|-----|------|---------|
| **Context-Engine (Jina-Code)** | **0.276** | 0.146 | 0.448 | 0.658 | 0.365 |
| Context-Engine (BGE-base) | 0.253 | 0.150 | 0.374 | 0.550 | 0.322 |
| CodeT5+ embedding | 0.266 | - | - | - | - |
| BM25 (Lucene) | 0.167 | - | - | - | - |
| BoW | 0.065 | - | - | - | - |
*Corpus: 20,604 code snippets | 500 queries | Pure dense retrieval, no reranking*
*Jina-Code: jinaai/jina-embeddings-v2-base-code (code-specific, 8k context)*
### CoIR Benchmark (Full Corpus, Dense Retrieval)
| Benchmark | Corpus | Queries | NDCG@10 |
|-----------|--------|---------|---------|
| **CodeSearchNet-Python** | 280K | 14.9K | **74.37%** |
| **CodeSearchNet-Go** | 280K | 14.9K | **74.51%** |
| **CodeSearchNet-JavaScript** | 280K | 14.9K | **57.19%** |
*Full CoIR corpus evaluation with dense retrieval (Jina-Code embeddings)*
---
## License
BUSL-1.1
