deepwiki

# DeepWiki MCP

Query any public GitHub repository using DeepWiki's AI-powered documentation and Q&A service. No API key, no auth, free.

**MCP endpoint:** `https://mcp.deepwiki.com/mcp`

## Scope & Boundaries

**This skill handles:**
- Asking natural-language questions about any public GitHub repo
- Listing documentation topics indexed by DeepWiki
- Fetching full wiki contents for a repo
- Running queries via the included helper script

**This skill does NOT handle:**
- Private repository access (requires paid Devin account)
- Modifying repositories or submitting PRs
- Real-time code analysis (DeepWiki may lag a few days behind latest commits)
- Local code search or grep (use standard file tools for that)

## Inputs

| Input | Required | Description |
|-------|----------|-------------|
| Question | Yes | Natural-language question about a repo |
| Repository | No | `owner/repo` format. Defaults to `openclaw/openclaw` |
| Action | No | `ask` (default), `topics`, or `docs` |

## Outputs

- AI-grounded text answer with source references from the repo
- Or a structured list of documentation topics
- Or full wiki contents (large output)

## Workflow

### Step 1 — Run the helper script

The script is located in this skill's directory at `scripts/deepwiki.sh`.

```bash
# Ask a question (defaults to openclaw/openclaw)
<skill_dir>/scripts/deepwiki.sh ask "How does session compaction work?"

# Ask about a specific repo
<skill_dir>/scripts/deepwiki.sh ask facebook/react "How does concurrent mode work?"

# List documentation topics
<skill_dir>/scripts/deepwiki.sh topics openclaw/openclaw

# Get full wiki contents (large output — prefer ask for targeted queries)
<skill_dir>/scripts/deepwiki.sh docs openclaw/openclaw
```

Replace `<skill_dir>` with the directory containing this SKILL.md.

### Step 2 — Interpret and relay the answer

DeepWiki returns AI-generated answers grounded in the repo's actual source code. The response typically includes:
- Direct answer to the question
- R...

# DeepWiki MCP

An [OpenClaw](https://openclaw.ai) skill that gives agents instant access to [DeepWiki](https://deepwiki.com), an AI-powered documentation and Q&A service for any public GitHub repository.

No API key. No auth. Free.

## What It Does

DeepWiki crawls public GitHub repos and builds a searchable, AI-grounded wiki. This skill exposes three capabilities via the [DeepWiki MCP server](https://mcp.deepwiki.com):

| Action | Description |
|--------|-------------|
| `ask` | Ask any natural-language question about a repo — get a sourced, grounded answer |
| `topics` | List all documentation topics DeepWiki has indexed for a repo |
| `docs` | Fetch the full wiki contents for a repo |

## Installation

Copy the skill directory to your OpenClaw workspace skills folder:

```bash
cp -r deepwiki ~/.openclaw/<workspace>/skills/
```

OpenClaw auto-discovers skills in `<workspace>/skills/`. Restart the gateway to activate.

### Requirements

- bash, curl, python3 (standard on most systems)
- No API keys or authentication needed

## Usage

The skill triggers automatically when asking about repository source code, architecture, or internals.

### Example prompts

- "How does session compaction work in OpenClaw?"
- "Ask DeepWiki about React's concurrent mode"
- "Look up the config schema in the codebase"
- "What's the architecture of torvalds/linux?"

### Helper script

```bash
# Ask a question (defaults to openclaw/openclaw)
scripts/deepwiki.sh ask "How does session compaction work?"

# Ask about a specific repo
scripts/deepwiki.sh ask facebook/react "How does concurrent mode work?"

# List documentation topics
scripts/deepwiki.sh topics openclaw/openclaw

# Get full wiki contents
scripts/deepwiki.sh docs openclaw/openclaw
```

## How It Works

DeepWiki exposes a [Model Context Protocol (MCP)](https://modelcontextprotocol.io) server at `https://mcp.deepwiki.com/mcp` using Streamable HTTP (JSON-RPC over SSE).

The helper script:
1. Builds the JSON-RPC payload
2. Posts to...