context6

Local code context indexer and retriever with:

• A CLI for indexing/search/lookup/snippets/summarization
• A FastMCP server exposing those capabilities as MCP tools

Attribution

This project is inspired by Context7.
Credit to the Context7 project and contributors for the original concept and direction.

What It Does

context6 scans a Python codebase, stores entities in SQLite, and lets you:

• Find symbols by name (lookup)
• Search by text (search)
• Pull exact source snippets (snippet)
• Generate compact summaries (summarize)
• Evaluate recall against qrels (eval-recall)

Prerequisites

• Python 3.10+
• pip (or another Python package installer)

Optional (only needed for summarization modes):

• Ollama running locally for ollama/auto mode
• OpenAI Codex CLI binary for codex mode fallback/direct use

Install

From repo root:

python -m pip install -e .

This installs the package and the context6 CLI entrypoint from pyproject.toml. If your shell cannot find context6, use module form: python -m context6.cli ....

Quick Start

1. Build an index for a source tree:

python -m context6.cli index --source /path/to/your/python/project --out ./arc

This creates ./arc/context6.db.

2. Search for relevant entities:

python -m context6.cli search --db ./arc/context6.db "database connection pooling"

3. Lookup by symbol:

python -m context6.cli lookup --db ./arc/context6.db --source /path/to/your/python/project MyClass

4. Get a snippet:

python -m context6.cli snippet --db ./arc/context6.db mypkg.module.MyClass.method_name

CLI Commands

Show all commands:

python -m context6.cli --help

Available subcommands:

• index: build an index and ingest into SQLite
• lookup: find best symbol match
• search: FTS search over fqname/signature/docstring/summary
• snippet: retrieve source text for a fully-qualified name
• summarize: fill missing summaries for indexed entities
• eval-recall: evaluate retrieval against qrels data

Useful examples

Filter by kinds:

python -m context6.cli search --db ./arc/context6.db "ARC execute" --kinds class,method

Run summarization:

python -m context6.cli summarize --db ./arc/context6.db --limit 100 --summarizer auto

Evaluate recall:

python -m context6.cli eval-recall --db ./arc/context6.db --qrels ./qrels.json --k 10 --retriever search

Environment Variables

These are used by CLI and/or MCP tools:

• CONTEXT6_SOURCE_ROOT: source root used when loading snippets
• CONTEXT6_DB: default DB path for MCP server (default: ~/code/context6/arc/context6.db)
• CONTEXT6_SUMMARIZER: auto, ollama, or codex for MCP summarize tool
• CONTEXT6_CODEX_BIN: path/name for codex binary (default: codex)
• CONTEXT6_MODEL: Ollama model (default: qwen2.5:7b)
• OLLAMA_HOST: Ollama endpoint (default: http://localhost:11434)
• CONTEXT6_NUM_CTX: Ollama context size override (default: 6144)

Running as an MCP Server

Run the server over stdio:

python -m context6.mcp.context6_server

Before launching, set at least:

export CONTEXT6_DB=/absolute/path/to/context6.db
export CONTEXT6_SOURCE_ROOT=/absolute/path/to/source/root

The server exposes tools including:

• context6_health
• context6_lookup
• context6_search
• context6_snippet
• context6_pretty_lookup
• context6_summarize_entity
• context6_resolve

Development

Run tests:

python -m pytest -q