What it covers
Building custom MCP (Model Context Protocol) servers so LLMs like Claude can read and act on systems they otherwise can’t reach, plus matching CLI tools so your team gets the same automation outside the chat. Recent build: ~50 tools and CLI commands for a B2B client, shipped from one codebase, with a SQLite mirror layer for big speedups on a slow upstream API.
Typical scope
Workshop to map which tools the LLM (and your team) actually need (less is more), TypeScript or Python MCP server with a dual MCP/CLI bin, deploy to your env or my managed setup, prompt-design guidance, and a doc that lets your team extend it later.
When to pick this
When you’ve tried RAG or generic chatbots and the model still doesn’t know enough about your domain. Or when you need scripted automation around a system whose UI or API gets in the way. MCP gives the model tools, the CLI gives the same tools to your team.
Deliverables
- MCP server in TypeScript or Python, tested with Claude Code
- Matching CLI so the same logic runs in scripts, cron, and CI
- Tool design: names, args, descriptions tuned for model use
- Auth, rate limiting, secret handling, deploy guide
- Documentation for end users and prompt-design notes