MCP servers & agent tooling
Make your product usable by Claude, ChatGPT and Copilot directly — not a chat widget bolted on the side. I build Model Context Protocol servers that expose your data and your actions as typed, permissioned tools an agent can actually call.
- ▸Python / TypeScript MCP SDK
- ▸Streamable HTTP + OAuth discovery
- ▸Bearer auth, hashed tokens, per-user scoping
- ▸Read + write tools with destructive-action guards
- ▸Hardened container: non-root, read-only, Vault secrets
- ▸Works as a remote connector or local stdio server
In production: a 34-tool MCP server over a Postgres project tracker — projects, tasks, notes, goals, reminders and calendar events, readable and writable from a Claude conversation, behind per-user token auth.







