contextcode 0.1.1

Code retrieval tool for coding agents

CodeContext

High-performance local MCP server that merges vector, lexical, and graph search into one retrieval engine for AI coding agents.

Quick Start • Features • Architecture • MCP Tools • Wiki • Install • CLI • Contributing

---

Stop guessing. Give your AI agents the precise codebase context they need to write production-grade code.

---

Quick Start

# 1. Install
pip install contextcode

# 2. Index your project (interactive config on first run)
cd /path/to/your/project
cctx index

# 3. Start the MCP server
cctx mcp

That's it. Your AI agent now has multi-stage retrieval across lexical, vector, and graph engines.

---

Features

Feature	Description
🔍 Multi-Stage Search	Lexical (FTS5) + Vector (ChromaDB) + Graph expansion + CrossEncoder reranking
🌲 AST-Aware Chunking	Tree-sitter parses 14 languages: Python, JS, TS, Go, Rust, Java, C/C++, C#, Dart, Bash, SQL, HCL, PowerShell, Markdown
🧠 Graph Intelligence	Dead code detection, call tracing, architecture overview, community detection
📖 Knowledge Wiki	LLM-maintained persistent wiki with OKF v0.1 compliance
🔗 Cross-Service Linking	HTTP route detection and inter-service call graph edges
⚡ GPU Embeddings	llama.cpp with Snowflake Arctic (Q8) for indexing, tiny model for queries
🔄 Incremental Indexing	Content-hash-based — only re-indexes changed files
🏗️ Pluggable Graph Store	Neo4j, AWS Neptune, DataStax Astra DB, and GraphRAG export adapters via a unified factory
🕸️ Static Call Graphs	Cross-language call extraction + import resolution, hub detection, and blast-radius analysis
📊 5-Signal Scoring	FTS + vector + name + signature + graph proximity composite ranking
🚀 Runtime Trace Ingestion	OTLP-style spans augment the static call graph with live data

---

Architecture

flowchart TD
    subgraph Ingestion["Ingestion Pipeline"]
        A[Codebase Files] --> B["Tree-sitter AST Chunker"]
        A --> C["Pattern Scanner"]
        A --> CX["Call Extractor + Import Resolver"]
        A --> D2["File Discovery<br/>.gitignore + .cctxignore"]
        B --> D[("SQLite FTS5<br/>Lexical Store")]
        B --> E[("ChromaDB<br/>Vector Store")]
        C --> F[("Graph Store<br/>Hierarchy Call Graph")]
        CX --> F
        F --> FB["GraphStoreFactory"]
        FB --> FN["Neo4j"]
        FB --> FP["AWS Neptune"]
        FB --> FD["DataStax Astra"]
        FB --> FG["GraphRAG Adapter"]
    end

    subgraph Retrieval["Retrieval Pipeline"]
        G[Agent Query] --> H["Parallel Dual Retrieval"]
        D -.-> H
        E -.-> H
        H --> I["RRF Fusion Ranker"]
        I --> J["Graph Expansion"]
        F -.-> J
        J --> K["CrossEncoder Reranker"]
        K --> L["High-Fidelity Context"]
    end

    subgraph Analysis["Graph Intelligence"]
        F --> M1["Dead Code Detection"]
        F --> M2["Call Path Tracing"]
        F --> M3["Architecture Overview"]
        F --> M4["Community Detection"]
        F --> M5["Git Diff Impact"]
    end

    subgraph Wiki["Knowledge Wiki"]
        L --> W1["Auto-sync on index"]
        W1 --> W2[("Markdown Wiki")]
        W2 --> W3["WikiQuery / WikiLint"]
    end

    style Ingestion fill:#0f172a,stroke:#38bdf8,stroke-width:2px,color:#fff
    style Retrieval fill:#0f172a,stroke:#34d399,stroke-width:2px,color:#fff
    style Analysis fill:#0f172a,stroke:#f59e0b,stroke-width:2px,color:#fff
    style Wiki fill:#0f172a,stroke:#a78bfa,stroke-width:2px,color:#fff

---

MCP Tools

CodeContext registers 11 MCP tools for AI agents:

Tool	Category	Description
CodeContext	Search	Multi-stage search: lexical + vector + graph expansion + reranking
WikiQuery	Wiki	Search the knowledge wiki for synthesized answers
WikiLint	Wiki	Health-check wiki for orphan pages, broken links, stale content
QueryGraph	Graph	Execute read-only Cypher queries against the knowledge graph
DeadCodeScan	Intelligence	Detect functions with zero callers (excluding entry points)
TracePath	Intelligence	BFS call tracing — who calls a function and what it calls (depth 1-5)
GetArchitecture	Intelligence	Architecture overview: languages, hotspots, entry points, modules
DetectChanges	Intelligence	Map git diff to affected symbols with blast radius and risk classification
DetectCommunities	Intelligence	Louvain community detection on the function call graph
LinkRoutes	Intelligence	Detect HTTP route handlers and create Route/HANDLES/HTTP_CALLS edges
IngestTraces	Intelligence	Ingest OTLP-style runtime traces to augment the static call graph

Usage Examples

# Multi-stage code search
cctx tools search "how does authentication work" --top-k 5

# Trace function call chains
cctx tools trace process_payment --direction outbound --depth 3

# Detect dead code
cctx tools dead-code

# Execute Cypher query
cctx tools query-graph "MATCH (f:Function)-[:CALLS]->(g:Function) RETURN f.name, g.name LIMIT 10"

# Architecture overview
cctx tools architecture

---

Wiki

CodeContext includes an LLM Wiki integration compliant with the Open Knowledge Format (OKF) v0.1. Instead of re-deriving knowledge on every query, the LLM builds and maintains a persistent wiki of interlinked markdown pages.

.codecontext/wiki/
  index.md              # OKF-compliant catalog
  log.md                # Update history
  modules/              # Module documentation
  patterns/             # Recurring patterns
  decisions/            # Architecture Decision Records
  concepts/             # Conceptual explanations

Page types: overview, module, pattern, decision, entity, comparison, concept, source, custom

---

Installation

Via pip

pip install contextcode

Via uv (recommended)

uv pip install contextcode

From Source

git clone https://github.com/VardhmanSurana/CodeContext.git
cd CodeContext
uv sync

Prerequisites

Requirement	Purpose	Required?
Python 3.11+	Runtime	✅ Yes
CUDA 12.x/13.x	GPU embedding acceleration	⚡ Recommended
Neo4j	Graph store (auto-starts if installed, unique constraints created on connect)	⚡ Recommended
AWS Neptune / DataStax Astra	Alternative managed graph backends	⬜ Optional

# Start Neo4j (optional — CodeContext also auto-starts it when installed)
sudo systemctl start neo4j
export NEO4J_PASSWORD='your-local-password'

Configure an alternative backend in .codecontext.yaml:

storage:
  graph:
    type: neptune        # neo4j | neptune | datastax | graphrag
    uri: wss://your-neptune-endpoint:8182/gremlin
    aws_region: us-east-1

---

Client Integration

Add to your MCP client config:

{
  "mcpServers": {
    "codecontext": {
      "command": "cctx",
      "args": ["mcp"]
    }
  }
}

Supported clients: Codex, Claude Desktop, Cursor, Windsurf, Hermes, OpenCode, Antigravity

Codex plugin

This repository includes a multi-client bundle at plugins/codecontext for Codex, Claude Code, Antigravity, and OpenCode. It exposes the CodeContext MCP server plus the codecontext and wiki-workflow skills. The bundle's MCP configurations are set for this source checkout; adjust the repository path when using a different checkout, then index the target project with cctx index before starting an agent.

---

CLI

Command	Description
cctx index	Index the current project
cctx index --rebuild	Force full rebuild
cctx index -v	Verbose output
cctx mcp	Start MCP server
cctx watch	Watch for file changes + auto-reindex
cctx wiki init	Initialize wiki directory
cctx wiki ingest	Ingest a file into the wiki
cctx wiki lint	Wiki health check
cctx tools search	Search codebase from CLI
cctx tools query-graph	Execute Cypher query
cctx tools trace	Trace function call chains
cctx tools dead-code	Detect dead code
cctx tools architecture	Architecture overview
cctx tools detect-changes	Git diff impact mapping
cctx tools communities	Community detection
cctx tools link-routes	Detect HTTP routes
cctx tools ingest-traces	Ingest runtime traces
cctx export	Export index for team sharing
cctx import	Import teammate's index
cctx graph migrate	Migrate the graph store to a new backend
cctx graph export	Export graph to Mermaid, D3, or interactive PyVis HTML
cctx graph viz	Run interactive call-graph visualization web server
cctx graph ingest	Ingest custom JSON code graphs

---

Graph Visualization & Exploration

The cctx graph command group turns the indexed call graph into shareable and interactive artifacts.

Export to multiple formats:

# Interactive PyVis HTML (default) — open in a browser to explore
cctx graph export -f html -o graph.html

# Mermaid diagram for docs
cctx graph export -f mermaid -o graph.md

# D3 JSON for custom dashboards
cctx graph export -f d3 -o graph.json

Live visualization server:

cctx graph viz --port 8765
# Open http://localhost:8765 and filter by search term or node type

Ingest custom graphs (e.g. runtime traces from another tool) as JSON:

{
  "nodes": [{ "labels": ["Function"], "properties": { "name": "main", "file_path": "app.py" } }],
  "edges": [{ "type": "CALLS", "source_node_id": "...", "target_node_id": "..." }]
}

cctx graph ingest -f graph.json

---

Contributing

1. Fork this repository
2. Create a feature branch: git checkout -b feat/my-feature
3. Make your changes and add tests
4. Run the test suite: pytest
5. Submit a Pull Request targeting feat/codecontext

Please report bugs and feature requests via GitHub Issues.

---

License

MIT — Crafted with care by the Advanced Agentic Coding Team.