Built for coding and code review agents · Self-hosted · MCP-native

ALLIGATOR Code Context Engine

Give your agents the what, where, and why of your codebase. Search code, explore dependencies, and understand the changes behind it. Persistent repository context for Codex and other clients through MCP.

OpenAI Built and evolved alongside Codex.
Hybrid Search
Dependency Graph
Commit Intelligence
MCP Native
Self-Hosted
Multi-Repo
Get CLI launch updates How it works // CLI coming to PyPI · Apache-2.0 licensed
What+Why
Not just where
11 tools
Core MCP tools
Hybrid
Dense + BM25
You own
Self-hosted data

// Core Capabilities

Context agents can actually use.

From the first investigation to the next code review. Bring code, relationships, and history into the same conversation.

⬡
Hybrid Code Search

Dense embeddings plus BM25 keyword search, fused for conceptual questions and exact symbol hits. Semantic when you need meaning; precise when you need a name.

⌬
Dependency Impact

Graph traversal over callers, callees, and imports. Ask what breaks if you change a function. Get a structured answer with source locations and relationship evidence.

◈
Temporal Intelligence

Commit history is first-class. Optional LLM summaries turn “who changed this” into “why it looks like this now” for agent investigations.

⟁
Smart Incremental Indexing

Detect changed files, update affected code, and reuse compatible parsing and embeddings. Keep context current with real-time file watching and Git URL synchronization.

⌾
MCP for Agents

11 core read-only tools for Codex and other clients through MCP. Search, investigate, retrieve source, and inspect index health.

⏣
Self-Hosted Stack

Qdrant + ArangoDB + Redis under your control. Your indexes stay on your infrastructure. Bring your own embedding and LLM providers.

⌗
Workspace Context

Search related repositories together. Follow supported cross-repository dependencies while retaining the repository, file, and source location of each result.

↗
Explore the Graph

Trace directed paths, inspect impact, and discover related file groups. Export a graph or explore it in an interactive HTML view.

↻
Resumable Indexing

Save progress as batches finish. Recover compatible work after an interruption and activate a completed generation for retrieval.

// Pipeline

From git repo to agent-ready context.

01
Index

Point Alligator at a local checkout or Git URL. Language-aware extraction builds searchable code and relationships in your own stack, with workspace support for related repositories.

TREE-SITTER · QDRANT · ARANGO
02
Enrich

Optionally summarize commits and embed history so agents can reason about intent, not only the latest snapshot of the code.

LLM SUMMARIES · COMMIT GRAPH
03
Serve

Run the MCP server over stdio or HTTP. Connect Codex and other clients in a few lines of configuration. Agents retrieve context from your published repository or workspace indexes.

FASTMCP · STDIO / HTTP
04
Investigate

Find an implementation, trace its dependencies, and inspect the changes behind it. Give coding and code review agents source-backed context in one flow.

SEARCH · GRAPH · HISTORY

// Code with context. Review with perspective.

Better questions. Grounded answers.

Ask Codex and other clients to use Alligator. Follow an answer back to its source.

Find

“Where is retry backoff implemented?”

Search by behavior or identifier, then retrieve the implementation with file and line attribution.

Review

“What depends on this function?”

Explore callers and dependency paths to give your code review the surrounding context.

Understand

“Why did this behavior change?”

Bring source history and optional commit summaries into the same investigation.

Your index runs on your infrastructure. Local models run locally; API-backed embeddings and summaries send selected content to your chosen provider.

// Early access

Be first to use the Alligator CLI.

Apache-2.0 licensed. Public CLI release coming to PyPI. Join the list for launch notes, installation guides, and release updates.

What you get at launch

A persistent context engine for Codex and other clients, running on infrastructure you control. Choose local or API-backed models to fit your workflow.

  • CLI distribution through PyPI
  • MCP setup for Codex and other clients
  • Architecture notes & indexing guide
  • Release notes and upgrade guidance

// Public CLI distribution. The source repository remains private.

No spam. Launch + major releases only. Unsubscribe anytime.

// Build status

Preparing the public CLI release.

$ alligator --help
context for coding and code review agents
$ alligator index ./my-monorepo --name platform
indexing… hybrid vectors · graph edges · commits
$ codex mcp add alligator -- alligator serve
Codex connection registered // stdio
tools: search_code · find_dependencies · investigate_code · …
status: private alpha

Private alpha. These bars track the named launch checks below, not a percentage of all possible features. Updated 18 September 2026.

Indexing + recovery75%
3 of 4 checks complete
  • Complete Language-aware extraction
  • Complete Incremental processing
  • Complete Checkpoint recovery validation
  • Pending Large-repository acceptance
Search + graph75%
3 of 4 checks complete
  • Complete Hybrid retrieval
  • Complete Graph and relationship workflows
  • Complete Live backend regression checks
  • Pending Large-repository retrieval verification
MCP tool surface100%
4 of 4 checks complete
  • Complete 11 core read-only tool definitions
  • Complete Repository and workspace scopes
  • Complete Graph CLI and MCP parity tests
  • Complete Client connection guides
CLI + operations75%
3 of 4 checks complete
  • Complete Managed backend setup
  • Complete Configuration and integrity diagnostics
  • Complete Durable job status
  • Pending Long-running deployment acceptance
Docs + installation75%
3 of 4 checks complete
  • Complete Installation guide
  • Complete First-index tutorial
  • Complete Codex connection guide
  • Pending Public PyPI installation verification
Public launch0%
0 of 2 checks complete
  • Pending Public CLI package published
  • Pending Public installation journey verified