AGENTS.md examples from real projects

nanobrowserReads credentialsAGENTS.md

nanobrowser

nanobrowser/nanobrowser/AGENTS.md

Nothing outside the extension can drive the layer or the agent.** There is no MCP for now. - **The conversation is compacted automatically before the context window fills**, and every message

BasedHardwareReads credentialsAGENTS.md

omi / backend

BasedHardware/omi/backend/AGENTS.md

MCP_OAUTH_PUBLIC_*` can expose a no-secret PKCE public client. Also set `MCP_AUTHORIZATION_SERVER_URL`, optional `MCP_RESOURCE_URL`, and token TTL env vars. - **llm-gateway** (`llm_gateway/main.py

tester-armyReads credentialsAGENTS.md

e2e

tester-army/e2e/AGENTS.md

src/agent/` the agent (the `act` executor socket plus the judgment methods), `src/mcp/` the `e2e mcp` server (a live session that rides the `act` socket with a queue executor so every

mozilla-firefoxAGENTS.md

firefox

mozilla-firefox/firefox/AGENTS.md

Firefox work - Some tools useful for Firefox work are available in the `moz` MCP server - Firefox is a very large repository, and it isn't efficient to search with usual

brokermr810Reads credentialsAGENTS.md

QuantDinger

brokermr810/QuantDinger/AGENTS.md

Repository map - `backend_api_python/`: Flask API, services, trading runtime, migrations, and backend tests. - `mcp_server/`: the tenant-scoped MCP client/server package backed by QuantDinger APIs. - `docs/`: product, architecture, deployment

doocsAGENTS.md

md

doocs/md/AGENTS.md

中执行命令 pnpm mcp # 在 @md/mcp-server 中执行命令（render_markdown 等 MCP 工具） pnpm mcp dev # MCP Server 监听模式 ``` `@md/mcp-server` 通过 stdio 暴露 `render_markdown`、`list_themes`、`list_colors` 等工具，配置见 [packages/mcp-server/README.md

darktable-orgAGENTS.md

darktable

darktable-org/darktable/AGENTS.md

when GTK3 has the Wayland backend gui/gtk.c calls into libwayland-client directly to detect server-side decoration support, but gtk+-3.0.pc does not list wayland-client. The symbols were

debeziumAGENTS.md

debezium

debezium/debezium/AGENTS.md

analyze the codebase, DO NOT set up build environments, DO NOT run scripts, or MCP, or skills, DO NOT look things up online, DO NOT make code changes, UNTIL

No other project uses it yet

formbricksReads credentialsAGENTS.md

formbricks

formbricks/formbricks/AGENTS.md

into `useState` or Jotai. The existing server actions wrap service calls and return `{ data }` or `{ error }` consistently; keep that contract when changing them. - MCP tools (`apps/web/modules/mcp`) are thin adapters over

openreplayAGENTS.md

openreplay / mcp_app

openreplay/openreplay/mcp_app/AGENTS.md

OpenReplay MCP App — Agent Guide Reference: https://modelcontextprotocol.github.io/ext-apps/api/documents/Patterns.html --- ## Project Structure ``` server.ts MCP server entry point (stdio transport) lib/ tools.ts All tool registrations (UI + internal) api.ts OpenReplay REST API client

VercelAGENTS.md

portless

vercel-labs/portless/AGENTS.md

change to a CLI command, flag, output, or environment variable must update the MCP tools in `packages/portless/src/mcp.ts` in the same change. `mcp.test.ts` reads `portless --help` and each tool command

No other project uses it yet

AutomatticReads credentialsAGENTS.md

wp-calypso

Automattic/wp-calypso/AGENTS.md

rendering primitives built with WordPress components. Agents Manager consumes it; conversation handling and MCP transport stay in consumers. - **Agenttic** (`packages/agenttic-client`, `packages/agenttic-ui`) — AI agent chat client library and UI components. Also

nearaiReads credentialsAGENTS.md

ironclaw

nearai/ironclaw/AGENTS.md

execution. - Sandboxed extension execution belongs in WASM or a runtime lane; external server integrations belong behind MCP and the network boundary. - Discovery is side-effect-free. Installation, credential binding, activation

nearaiAGENTS.md

ironclaw / tests

nearai/ironclaw/tests/AGENTS.md

before durable storage without retaining inline media/blob base64 | `mcp.rs` | | User-registered and bundled hosted MCP servers register, authenticate, project active, restore, and invoke | `hosted_mcp_registration.rs` | | Web search/fetch runs the real

tw93AGENTS.md

Kami

tw93/Kami/AGENTS.md

WeasyPrint-safe. Pure Python, no Node, ships in the package. - `scripts/mcp_server.py` - zero-dependency MCP stdio server exposing `kami_templates` / `kami_doctor` / `kami_render` / `kami_check` / `kami_deliver` / `kami_screenshot

ChainlitAGENTS.md

chainlit

Chainlit/chainlit/AGENTS.md

over keeping legacy code in place. ## MCP-First Approach (CRITICAL) When available, **ALWAYS** prefer MCP servers over manual alternatives. Use **Context7** for docs/API references, **Serena** for code navigation/refactoring/memory, and **GitHub

MemTensorReads credentialsAGENTS.md

MemOS

MemTensor/MemOS/AGENTS.md

parsers/` `reranker/` | Provider implementations (`base.py` + `factory.py` + each backend) | | `src/memos/api/` | FastAPI service (routers / handlers / middleware / MCP server) | | `src/memos/configs/` | All pydantic configuration classes (one-to-one with the modules above) | | `src/memos/context/` | Cross

jo-incReads credentialsAGENTS.md

camofox-browser

jo-inc/camofox-browser/AGENTS.md

Camofox has two front doors: the HTTP API and the agent tool interface (an MCP server, wrapped by the OpenClaw plugin). **Every tool-like browser action must be reachable from

viarotel-orgAGENTS.md

escrcpy

viarotel-org/escrcpy/AGENTS.md

string | string[]` contract; business code stays untouched across internal changes. - [packages/madb/](../packages/madb/) is an MCP server for AI agent Android device control via ADB and yadb (32 tools

mesheryReads credentialsAGENTS.md

meshery

meshery/meshery/AGENTS.md

navigate, snapshot, click, fill forms, run JS, inspect console/network) in place of raw Playwright/chrome-devtools MCP for ad hoc tasks. - Run `quota-axi` to check local agent-provider quota windows before

No other project uses it yet