Claw Skill

Claw Skill exposes GsKnora to agents in the OpenClaw ecosystem via the REST API: send documents, import web pages, and search across knowledge bases.

The skill is hosted on ClawHub as @lyingbug/weknora, MIT-0 license. Effective capabilities depend on the connected GsKnora API.

Main features

Capability	Matching interface
File upload	PDF, Word, Excel, etc. to a knowledge base with parsing and vectorization
Web import	Capture by URL, write to a knowledge base, track parse status
Markdown write	Create or edit Markdown entries (reports, structured notes)
Hybrid search	Single-base hybrid-search and multi-base knowledge-search (vector + keywords)
Browse knowledge bases	List bases and entries, view detail

Install and connect

Under Settings → Publish integrations → Claw Skill, copy the API URL, sample environment variables, and install command.

1. API credentials: Settings → Publish integrations → API integration (Open API info button);

2. Environment variables: terminal or ~/.zshrc / ~/.bashrc

   export WEKNORA_BASE_URL=https://your-weknora.example.com/api/v1
   export WEKNORA_API_KEY=sk-xxxxx

   WEKNORA_BASE_URL must include /api/v1; the guide example uses the current instance address.

3. Install the skill: with OpenClaw CLI, or via ClawHub;

   openclaw skills install @lyingbug/weknora

4. Verify: ask the agent to list knowledge bases or run a search.

Relation to MCP

Claw Skill and the MCP server both let external agents call GsKnora; choose based on client and desired features:

	Claw Skill	MCP server (built-in)
Target	OpenClaw / ClawHub agents	MCP clients (Claude Desktop, Cursor, Claude Code, VS Code Copilot, etc.)
Install	ClawHub skill	No extra deploy: create an endpoint under Settings → Publish integrations → MCP server, connect at /mcp/<endpoint_id>
Auth	Workspace API key (WEKNORA_API_KEY)	Per-endpoint token, rotation and revocation
Transport	Direct REST	Streamable HTTP; STDIO fallback mcp-remote@0.15.1 when no native HTTP/Bearer
Scope	Import, search, browse (5 types)	Tools selected per endpoint: search/read, Q&A (default agent), Wiki, write; knowledge bases can be limited
Docs	This document	MCP integration

For Q&A, Wiki tools, or fine-grained tool/base limits, prefer the MCP server; for import/search/browse in OpenClaw, Claw Skill is enough. The Python MCP service under mcp-server/ is deprecated; new integrations use the built-in MCP server.

Related documentation

• Credentials and capability restriction: Tenants, users, and authentication
• Underlying interfaces: API overview
• Other integrations: Chrome extension, MCP integration