AI integration and MCP
Connect AI assistants to your product and your data, with the access controls, reliability and cost discipline of any other production system.
The Model Context Protocol (MCP) is the standard way for AI assistants such as Claude to work with other software. An MCP server lets an assistant search, read and act on your product on a user’s behalf. Done well, it makes your product useful from inside the tools people already use. Done badly, it’s a security hole.
What I build
- MCP servers for your product or platform, with authentication, carefully scoped tools, and limits on what an assistant is allowed to change.
- CLIs and developer tooling that people and AI agents can both drive.
- LLM features in your own application: streaming interfaces, resilient APIs, and keeping latency, correctness and cost under control.
- Agent workflows for your engineering team, so coding agents can help without being handed the keys.
I’ll also tell you when AI isn’t the right tool for the job.
Experience
- caffeine.ai: designed and built Caffeine’s CLI and its MCP server from a blank page, on my own, and took the MCP server through to a listing in Anthropic’s Claude directory.
- An AI app-building platform: a local-first migration agent in which AI only ever sees data the user has reviewed, redacted and approved, with a local MCP server and a hash-chained audit trail of every disclosure.
- Rilbo: our own project tracker includes an MCP server, with scoped tools and a per-session switch that turns off write access.