How real projects brief Claude Code: the commands, conventions and warnings they put in CLAUDE.md.
ThinkWatchProjectCLAUDE.mdThinkWatchProject/ThinkWatch/CLAUDE.md
Project structure - `crates/server` — Dual-port Axum server (gateway :3000 + console :3001) - `crates/gateway` — AI API proxy logic (OpenAI, Anthropic, Gemini, Azure, Bedrock) - `crates/mcp-gateway` — MCP tool proxy - `crates/auth` — JWT, API key, OIDC
AVIDS2CLAUDE.mdAVIDS2/memorix/CLAUDE.md
skip project memory just because the initial MCP status says `pending` Claude Code starts MCP servers asynchronously. An initial `pending` MCP status is not a failure by itself; the tools
liangdabiaoCLAUDE.mdliangdabiao/amazon-sorftime-research-MCP-skill/CLAUDE.md
Claude skills ## Sorftime MCP Integration ### API Configuration - **Config file**: `.mcp.json` - stores API key and endpoint - **Endpoint**: `https://mcp.sorftime.com?key={API_KEY}` - **Protocol**: `streamableHttp` using Server-Sent Events
StripeCLAUDE.mdstripe/link-cli/CLAUDE.md
take`, `identity credentials request|list|present`, `report`, `serve`. The CLI also runs as an MCP server (`--mcp`) and serves skill files via `skills` subcommand, both provided by incur. **When changing
MicrosoftCLAUDE.mdmicrosoft/CopilotStudioSamples/CLAUDE.md
authoring/ # Importable solutions and topic snippets ├── contact-center/ # ServiceNow, Salesforce, Genesys, skill handoff ├── extensibility/ # MCP servers, A2A protocol, M365 Agents SDK ├── guides/ # Implementation guide, workshop, playbook ├── infrastructure/ # VNet and deployment
dark-hxxCLAUDE.mddark-hxx/CLI-Manager/CLAUDE.md
none/running/attention/done/failed）。改动状态逻辑时要同时考虑两个来源。 ### CLI Hook 桥接（Claude / Codex） - `src-tauri/src/claude_hook.rs` 在启动时绑定 `127.0.0.1` 随机端口起一个 TCP server，用一次性 token 校验，接收 Claude/Codex 的 hook 上报（SessionStart/UserPromptSubmit/Notification/Stop/StopFailure/PermissionRequest），转成 `claude-hook-notification` 事件发给前端。 - `hook_settings.rs
LangChainCLAUDE.mdlangchain-ai/deep_research_from_scratch/CLAUDE.md
clarification and brief generation │ ├── 2_research_agent.ipynb # Research agent with custom tools │ ├── 3_research_agent_mcp.ipynb # Research agent with MCP servers │ ├── 4_research_supervisor.ipynb # Multi-agent supervisor coordination │ ├── 5_full_agent.ipynb # Complete end-to-end system │ └── utils.py # Shared utilities
CodeBendKitCLAUDE.mdCodeBendKit/codeseek/CLAUDE.md
success: bool` and `data: T`. ## Configuration Config file at `~/.codeseek/config/config.toml`. Key sections: - `[http]` — `server_port`, `codebase_port` (default 12800) - `[codebase]` — `enable_embedding`, `embedding_db_uri`, `graph_db_uri` - `[codebase.embedding]` — `model
ethanplusaiCLAUDE.mdethanplusai/jarvis/CLAUDE.md
user's Claude subscription, fed over stdin as stream-json - `jarvis_mcp.py` — Stdio MCP server exposing JARVIS's tools to the brain; forwards `tools/call` to `POST /internal/tool` - `speech.py` — Sentence splitting
ethanplusaiCLAUDE.mdethanplusai/jarvis/jarvis_home/CLAUDE.md
Markdown files he can open, not rows in a database. - He brings his own MCP servers; you ship connected to nothing. - 1,651 tests against the public repo
jwangkunCLAUDE.mdjwangkun/claude-for-financial-services-cn/CLAUDE.md
vendored copies of their skills. **Never edit bundled copies directly.** ## Core data layer Four MCP servers (multi-tier architecture): ```bash # Start the Wind data server (万得, paid — requires WIND
MicrosoftCLAUDE.mdmicrosoft/TypeAgent/ts/examples/planValidationDemo/CLAUDE.md
Plan-Validated Agent Execution This project is governed by a plan-validation MCP server. **Every file access and shell call you make must go through the `validated_*` tools.** The built
symgraphCLAUDE.mdsymgraph/GhidrAssistMCP/CLAUDE.md
Context Protocol support - Jetty 11.0.20 for embedded HTTP server - Jackson 2.17.0 for JSON processing ### MCP Server The extension includes an embedded MCP (Model Context Protocol) server that: - Runs on port
tbxarkCLAUDE.mdtbxark/mcp-proxy/CLAUDE.md
MCP proxy server (Go, single `main` package, flat layout) that aggregates multiple downstream MCP servers behind one HTTP entrypoint. Each configured downstream server is exposed at its own route
GarethManningCLAUDE.mdGarethManning/education-agent-skills/CLAUDE.md
Claude Code, Claude.ai (via MCP), OpenAI Codex, and Hermes Agent. Includes a live MCP server at mcp-server-sigma-sooty.vercel.app/mcp, a CoWork plugin (.claude-plugin/), a Codex plugin (.codex-plugin/), and Hermes
kgoedeckeCLAUDE.mdkgoedecke/doop/CLAUDE.md
sandboxed iframes. People edit in the browser; AI agents edit through the built-in MCP server, streaming designs in live. Everyone sees cursors, presence, frame edits, agent status
ronak-createCLAUDE.mdronak-create/FableCut/CLAUDE.md
drive the editor. ## MCP connection (preferred — works from any session, any directory) Register the MCP server (`mcp-server.js`) once at user scope as `fablecut`: `claude mcp add -s user fablecut -- node
codewithmukeshCLAUDE.mdcodewithmukesh/dotnet-claude-kit/CLAUDE.md
companion for .NET developers. It provides skills, agents, templates, knowledge documents, and a Roslyn MCP server that make Claude Code dramatically more effective for .NET development. ## Philosophy - **Guided over prescriptive
borskiCLAUDE.mdborski/travel-hacking-toolkit/CLAUDE.md
Plans trips with points, miles, awards, and cash. Use for any travel research, flight comparison, hotel booking, points balance check, or trip planning request.
sandbox-quantumCLAUDE.mdsandbox-quantum/switch/CLAUDE.md
clients/` — room clients (agent, admin, bridge) - `bridges/` — External integrations - `agent/` — Agent Bridge (HTTP API, MCP server, server-side connectors) - `collaboration/` — Collaboration Bridge (Slack, Mattermost, Discord, Teams, Telegram adapters) - `resource/` — Resource
A file Claude Code reads at the start of every session. It holds the commands, conventions and warnings the agent needs for this project.
At the repository root. Claude Code also reads CLAUDE.md files in subdirectories when it works there.
Build and test commands, the project's layout, conventions that aren't obvious from the code, and mistakes to avoid. Short files tend to work better than long ones.
Claude Code reads CLAUDE.md; most other agents read AGENTS.md. Many projects keep one and point the other at it.
