deepwiki-mcp（deepwiki MCP 搜索）

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

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.

# 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
• References to specific files and code paths
• Context about related functionality

Relay the answer to the user, adding your own context if you have additional knowledge.

Step 3 — Follow up if needed

If the answer is incomplete or raises new questions:

• Ask a more specific follow-up question
• Use topics to find relevant documentation sections
• Use docs for broader context (but note: output can be very large)

Direct curl (fallback)

If the helper script is unavailable:

curl -s -X POST https://mcp.deepwiki.com/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "ask_question",
      "arguments": {
        "repoName": "owner/repo",
        "question": "YOUR QUESTION"
      }
    }
  }' | grep '^data:' | grep '"id":1' | sed 's/^data: //' | \
  python3 -c "import json,sys; d=json.load(sys.stdin); print(d['result']['content'][0]['text'])"

MCP Tools Reference

Error Handling

Success Criteria

• DeepWiki returns a substantive answer (not an error or empty response)
• Answer references actual code/files from the repository
• User's question is addressed with grounded information

Configuration

No persistent configuration required. The skill uses:

• exec tool to run the helper script (bash + curl + python3)
• No API keys or authentication needed
• Works for any public GitHub repository

System dependencies:

Notes

• Responses take 10-30s (AI generates answers server-side)
• ask_question is the most useful tool — use it first
• DeepWiki crawls repos periodically; may lag behind very recent commits
• Works for any public GitHub repo, not just OpenClaw
• For private repos, a paid Devin account is required