Relay
Enables AI agents to access a unified catalog of tools from various APIs (OpenAPI, GraphQL, MCP, Google Discovery) through the MCP protocol.
README
<p align="center"> <img src="assets/relay-icon.png" alt="Relay" width="80" /> </p>
<h1 align="center">Relay</h1>
<p align="center"> <strong>The integration layer for AI agents.</strong><br/> One catalog for every tool, shared across every agent you use. </p>
<p align="center"> <a href="https://www.npmjs.com/package/relay"><img src="https://img.shields.io/npm/v/relay?color=3B82F6&label=npm&style=flat-square" alt="npm version" /></a> <a href="https://github.com/gothamdev244/relay/blob/main/LICENSE"><img src="https://img.shields.io/github/license/gothamdev244/relay?color=10B981&style=flat-square" alt="license" /></a> <a href="https://discord.gg/eF29HBHwM6"><img src="https://img.shields.io/badge/discord-join-5865F2?style=flat-square&logo=discord&logoColor=white" alt="discord" /></a> </p>
<p align="center"> <img src="assets/demo.gif" alt="Relay CLI Demo" width="600" /> </p>
Relay is an open-source runtime that unifies your integrations — OpenAPI, GraphQL, MCP, and Google Discovery — into a single tool catalog. Connect once, use everywhere: from Claude Code and Cursor to custom agents built with the SDK.
Architecture
<p align="center"> <img src="assets/architecture.svg" alt="Relay Architecture" width="100%" /> </p>
Quick Start
npm install -g relay
relay web
Opens a local runtime with a web UI at http://127.0.0.1:4788. Add your first source and start using tools.
Use as an MCP Server
Point any MCP-compatible agent at Relay to share your tool catalog, auth, and policies across all of them.
relay mcp
Example mcp.json for Claude Code / Cursor:
{
"mcpServers": {
"relay": {
"command": "relay",
"args": ["mcp"]
}
}
}
SDK
Three entry points — pick the one that fits:
| Import | Use case |
|---|---|
@relay-sh/sdk |
Promise-based. Scripts, quick integrations, non-Effect codebases. |
@relay-sh/sdk/core |
Effect-native. Full type safety, streaming, composable pipelines. |
@relay-sh/sdk/testing |
Mocks and utilities for writing tests against your tools. |
import { createRelay } from "@relay-sh/sdk";
const relay = await createRelay();
await relay.openapi.addSpec({
url: "https://petstore3.swagger.io/api/v3/openapi.json",
});
const tools = await relay.tools.list();
console.log(tools);
await relay.close();
Adding Sources
Relay has first-party support for OpenAPI, GraphQL, MCP, and Google Discovery. The plugin system is open to any source type.
Web UI
Open http://127.0.0.1:4788 → Add Source → paste a URL. Relay detects the type, indexes tools, and handles auth.
CLI
relay call relay openapi addSource '{
"spec": "https://petstore3.swagger.io/api/v3/openapi.json",
"namespace": "petstore",
"baseUrl": "https://petstore3.swagger.io/api/v3"
}'
Using Tools
Programmatic
const matches = await tools.discover({ query: "github issues", limit: 5 });
const detail = await tools.describe.tool({
path: matches.bestPath,
includeSchemas: true,
});
const issues = await tools.github.issues.list({
owner: "vercel",
repo: "next.js",
});
CLI
relay tools search "send email"
relay call github issues create '{"owner":"octocat","repo":"Hello-World","title":"Hi"}'
relay call gmail send '{"to":"alice@example.com","subject":"Hi"}'
If an execution pauses for auth or approval:
relay resume --execution-id exec_123
CLI Reference
relay web Start runtime + web UI
relay daemon run | status | stop Manage background daemon
relay mcp Start MCP endpoint
relay call <path...> '{"k":"v"}' Invoke a tool by path
relay call <path...> --help Browse namespaces and methods
relay resume --execution-id <id> Resume paused execution
relay tools search "<query>" Search tools by intent
relay tools sources List sources + tool counts
relay tools describe <path> Show tool schema
Project Structure
apps/
cli/ CLI (relay command)
local/ Local runtime server
cloud/ Hosted service
desktop/ Desktop app
marketing/ relay.sh website
packages/
core/
sdk/ Core contracts, plugin wiring, scopes, policies
execution/ Execution engine
config/ Configuration and scope management
storage-*/ Storage adapters
plugins/
openapi/ OpenAPI plugin
graphql/ GraphQL plugin
mcp/ MCP plugin
google-*/ Google Discovery plugin
kernel/
quickjs/ QuickJS sandboxed runtime
deno/ Deno subprocess runtime
ir/ IR compiler
react/ React UI components
hosts/
mcp/ MCP host surface
Development
bun install
bun run dev
Dev server starts at http://127.0.0.1:4788.
bun run test # run tests
bun run lint # oxlint
bun run typecheck # type checking
bun run format # oxfmt
See CONTRIBUTING.md for full setup instructions.
License
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。