Credentials. The stdio server reads LLM_* credentials from the environment that the agent launches it with.
Who it is forA developer working in Claude Code who wants project knowledge from compiled notes available without leaving the editor.
The problemCoding agents have no access to the decisions and context stored in your personal or team notes.
The outcomeClaude Code gets a single ask tool that returns a cited Markdown answer grounded in the wiki.
What you need first
- An OpenAI-compatible LLM API key and endpoint (OpenAI, DeepSeek, Ollama, vLLM, Azure OpenAI, etc.)
- Docker, or a Linux amd64/arm64 or macOS arm64 machine for the CLI install
- Your source files, notes, or URLs to compile
Steps
- Build or install a knowledge base as described in the Quick Start so the wiki exists on disk.
- Set
KAAS_KB_DIRto the knowledge-base root and theLLM_*credentials in your environment. - Register the server with
claude mcp add kaas -- kb-ai mcp. - Ask Claude Code a question that the wiki should answer and check that the reply cites wiki articles.
Prompt for your agent
Using the KaaS repo at bybit-exchange/kaas, register the stdio MCP server for the knowledge base at <kb-dir> with Claude Code. Expected output: a Claude Code session where the ask tool returns a cited Markdown answer. Verify that the cited article path exists under <kb-dir>.
Caveat. The README does not say how to install the kb-ai command on its own, so confirm it is on your PATH before registering it.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for bybit-exchange/kaas and "Let Claude Code query your compiled wiki locally"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer