High-performance code intelligence MCP server. Indexes codebases into a persistent knowledge graph — average repo in milliseconds. 158 languages, sub-ms queries, 99% fewer tokens. Single static binary, zero dependencies.
MCP gateway for shared agent tools and skills
AAI Gateway sits between your agents and the tools or skills they use. It can discover, import, enable, disable, and remove Agent Apps, then expose each one as a tool your agent can call.
Builders who want Claude Code, Codex, or other agents to share the same installed tools and skills.
You can set up a tool or skill once and use it across agents without repeating config or re-explaining context.
What it does
Search and install Agent Apps
`search:discover` finds tools and skills with natural language and installs them.
Import MCP servers and skills
`mcp:import` and `skill:import` turn external MCP servers and skill packages into Agent Apps.
Manage apps per agent
`listAllAaiApps`, `enableApp`, `disableApp`, and `removeApp` control what each agent can use.
Expose imported apps as tools
Each imported Agent App becomes an `app_<app-id>` tool with its guide and tool list.
Auto-discover preset apps
Local installs can auto-discover preset apps like `claude`, `codex`, and `opencode`.
How to get it
- 1Claude Code
claude mcp add --scope user --transport stdio aai-gateway -- npx -y aai-gateway
- 2Codex
codex mcp add aai-gateway -- npx -y aai-gateway
- 3OpenClaw
openclaw plugins install openclaw-aai-gateway-plugin
README
AAI Gateway: Unified MCP & Skill Management, Shared Across AI Agents, 99% Context Token Savings
What Is It
AAI = Agent App Interface
AAI Gateway is the interaction gateway for Agent Apps.
What is an Agent App? An Agent App is a collection of capabilities that an Agent can use. For example:
- An MCP Server is an Agent App — it provides a set of tools
- A Skill package is an Agent App — it provides one or more skills
In AAI Gateway, they are abstracted as Agent Apps under unified management. Import once, and all AI Agents can use them immediately.
What Problems Does It Solve
Context Bloat
Traditional: 10 MCPs × 5 tools = 50 full schemas ≈ 7,500 tokens injected into every conversation.
AAI Gateway: each Agent App needs only fewer than 50 tokens for a summary, with details loaded on demand. 99% token savings.
Finding Tools Is Hard
Traditional: search GitHub → read READMEs → copy JSON configs → debug connections → restart Agent.
AAI Gateway: tell your Agent "use AAI to search for xxx" — auto-searches, installs, ready to use.
"Use AAI to search for a browser automation tool"
→ Search → finds Playwright MCP → Agent summarizes a one-line Agent App summary → install → ready to use, no restart needed
"Use AAI to search for a PPT creation skill"
→ Search → finds PPT Skill → uses skill description as Agent App summary → install → ready to use, no restart needed
Duplicate Config
Configure the same thing in Claude Code, Codex, and OpenCode separately? Import once through AAI Gateway, all Agents share instantly.
Quick Start (30 Seconds)
Claude Code:
claude mcp add --scope user --transport stdio aai-gateway -- npx -y aai-gateway
Codex:
codex mcp add aai-gateway -- npx -y aai-gateway
OpenCode — add to ~/.config/opencode/opencode.json:
{
"mcp": {
"aai-gateway": {
"type": "local",
"command": ["npx", "-y", "aai-gateway"],
"enabled": true
}
}
}
openclaw plugins install openclaw-aai-gateway-plugin
Once installed, just tell your Agent what you want to do.
Built-in Tools
| Tool | Description |
|---|---|
search:discover | Search and install new tools with natural language |
mcp:import | Import an MCP Server as an Agent App |
skill:import | Import a Skill package as an Agent App |
listAllAaiApps | List all registered Agent Apps |
enableApp / disableApp | Enable or disable an Agent App per Agent |
removeApp | Remove an Agent App |
aai:exec | Execute a specific tool within an Agent App |
Each imported Agent App generates an app_<app-id> tool that returns the full operation guide and tool list when called.
Preset Agent Apps (auto-discovered when locally installed)
| App ID | Name | Description |
|---|---|---|
claude | Claude Code | AI coding assistant for code editing, analysis, and development |
codex | Codex | OpenAI-powered AI coding assistant |
opencode | OpenCode | AI development assistant for editing files and running commands |
Architecture

Developers: Get Your Agent App Auto-Discovered
Create an aai.json descriptor and submit it to src/discovery/descriptors/. When a user's local environment meets the discovery.checks conditions, the Agent will auto-discover your Agent App.
{
"schemaVersion": "2.0",
"version": "1.0.0",
"app": {
"name": { "default": "My App" }
},
"discovery": {
"checks": [
{ "kind": "command", "command": "my-app" }
]
},
"access": {
"protocol": "mcp",
"config": {
"command": "my-app-mcp",
"args": ["--stdio"]
}
},
"exposure": {
"summary": "Use when the user wants to do X."
}
}
discovery.checks supports three check types: command (command exists), file (file exists), path (directory exists).
Supported protocols: mcp, skill, acp-agent
Welcome to submit a PR to contribute new Agent App descriptors, or open an issue for feedback.
Files in the repo
- docs
- images
- src
- .gitignore
- .prettierrc
- CLAUDE.md
- CONTRIBUTING.md
- eslint.config.mjs
- LICENSE
- package-lock.json
- package.json
- README.ja.md
- README.ko.md
- README.md
- README.zh-CN.md
- tsconfig.json
- vite.config.ts
- vitest.config.ts
Discussion (0)
Ask about usage, or say what you built with itSign in to join the discussion.
No comments yet. Be the first to say what this is good for.
More connectors

Universal provider proxy for OpenAI Codex & Claude Code — use any LLM (Claude, Gemini, Grok, DeepSeek, Ollama…) with Codex CLI, App, SDK, and Claude Code
Git-native persistent memory for AI coding agents. Implements Google OKF v0.2 with sub-300µs in-memory BM25 search, embedded MCP server, and progressive disclosure. Slashes token bloat by 80% with zero external databases or dependencies. Built in pure Go.
Local-first code intelligence graph for MCP and CLI. Builds a persistent map of your codebase so AI coding tools read only what matters, with benchmarked context reductions on reviews and large-repo workflows.
Stop your AI from making things up — it proposes, deterministic tools decide, every claim checked against ground truth with evidence. Grounded facts and context survive resets. Reverse engineering is the proving ground. MCP server + CLI.
20 MB lightweight cross-platform database client for 90+ databases, including MySQL, PostgreSQL, SQLite, Redis, MongoDB, DuckDB, SQL Server, and Dameng. Built-in AI, MCP Server, CLI, desktop and Docker. | 轻量级跨平台数据库管理工具,支持 MySQL、PostgreSQL、SQLite、Redis、MongoDB、达梦等 90+ 数据库,提供桌面端、Docker、CLI、内置 AI 助手和 MCP Server。