eos

<div align="center"> # EOS — GitHub Org Context API **The context layer for AI across your entire GitHub organisation.** Point it at your org and any agent understands it — who owns what, how the code is structured, who to ask, what's changing, and how work flows — in one call. [](https://www.npmjs.com/package/@eos-ai/sdk) [](https://www.npmjs.com/package/@eos-ai/mcp-server) [](./LICENSE) [Docs](https://www.eos.dev/developers/api-reference) · [Get an API key](https://www.eos.dev/settings/api) · [eos.dev](https://www.eos.dev) </div> --- Every AI agent, coding tool, and internal workflow is only as good as its context about your organisation. Today each tool reconstructs that from the raw GitHub REST/GraphQL API: rate-limited, shallow, and rebuilt from scratch every time. The **GitHub Org Context API** is a single endpoint that turns your whole org into structured, LLM-ready context — ownership, dependencies, expertise, review history, architecture, and activity — served over REST, an official **MCP server**, and TypeScript/Python SDKs. No crawling. No rate limits. No rebuilding the same graph in-house. ## Quickstart Grab a key at [eos.dev/settings/api](https://www.eos.dev/settings/api), then: ```bash npm install @eos-ai/sdk ``` ```ts import { EosClient } from "@eos-ai/sdk"; const eos = new EosClient({ apiKey: process.env.EOS_API_KEY }); // Who owns this part of the codebase? const { data } = await eos.context.ownership({ repo: "acme/api", path: "src/auth" }); for (const owner of data) { console.log(`${owner.handle} — ${owner.commits} commits`); } ``` That's it — you're in production against your org's context. ## Connect it to Claude / Cursor / any MCP client Add the official MCP server and your agent gains org context as native tools — zero glue code: ```jsonc { "mcpServers": { "eos": { "command": "npx", "args": ["-y", "@eos-ai/mcp-server"], "env": { "EOS_API_KEY": "eos_your_key_here" } } } } ``` ## Context surfaces | Surface | What it returns | |---|---| | `context.ownership` | Who owns/knows a file, module, or system (+ bus-factor signal) | | `context.graph` | Module/dependency graph — repos, modules, co-change coupling | | `context.contributors` | Per-person expertise and recent activity | | `context.pulls` | PRs in flight — the review graph, turnaround, who's waiting | | `context.search` | Natural-language retrieval → LLM-ready chunks **with citations** | | `context.activity` | Hotspots — where change is concentrated right now | Every response is the standard `{ data, meta }` envelope, rate-limit-free and pre-computed. Lists are cursor-paginated; the SDK's `client.paginate()` follows them for you. ## What's in this repo | Package | Description | |---|---| | [`packages/sdk`](./packages/sdk) — `@eos-ai/sdk` | Typed TypeScript client for every surface, with retries + pagination | | [`packages/mcp-server`](./packages/mcp-server) — `@eos-ai/mcp-server` | Official Model Context Protocol server | | [`packages/cli`](./packages/cli) — `@eos-ai/cli` | `eos` command-line tool | | [`examples`](./examples) | Runnable integration examples | A Python SDK (`eos-context`) is on the way. ## Why not just call the GitHub API? GitHub gives you **events**; EOS gives you **context** — semantic, pre-computed, relationship-aware, rate-limit-free, and LLM-ready. It models the *organisation* (people, ownership, dependencies, flow), not just code text, and it's the shared, maintained layer instead of undifferentiated in-house heavy lifting. Coding assistants aren't competitors — EOS is the context layer *underneath* them, making their agents smarter about *your* org. ## Contributing Issues and PRs welcome — see [CONTRIBUTING.md](./CONTRIBUTING.md) and our [Code of Conduct](./CODE_OF_CONDUCT.md). ## License [MIT](./LICENSE) © EOS