MCP Bridges
A growing set of custom MCP servers that let Claude read and operate our real infrastructure — DNS, LaTeX docs, git, trading, and household finance — from inside a single conversation.
The Idea
Most people use Claude to write text or answer questions. The moment it can also read your DNS records, push a commit, compile a LaTeX document, or query a trading portfolio, the conversation stops being a chat and becomes a console for your whole stack.
That's what the Model Context Protocol actually buys you: a standard way to expose a system's real actions as tools an LLM can call directly, instead of Claude guessing at shell commands or drafting instructions for a human to type in by hand. We build one of these bridges per service, and every one of them turns something we used to click through into something we can just ask for.
The Bridges We've Built
DNS
Add, update, or delete records directly from chat. No more logging into a registrar dashboard to point a subdomain — just ask.
Overleaf
Read and write LaTeX projects and trigger compiles without opening the editor. Draft a document, cite a source, see the compiled PDF, in one conversation.
Gitea
Open issues, review pull requests, cut releases, and drive CI/CD on our self-hosted git server — the full software lifecycle without leaving chat.
Ledger & Portfolio Tools
Purpose-built tools onto our household ledger and personal trading systems let Claude check a balance, categorize a transaction, or post a journal entry the same way it edits code — domain-specific actions, not read-only exports. The ledger has its own page elsewhere in this portfolio.
Vivid
Our self-hosted image generator (its own project, built end-to-end by Claude) also ships an MCP endpoint, so Claude can commission original art the same way it queries a database.
Why This Matters
The interesting part isn't any single bridge — it's that Claude wrote most of them. We describe what a service is capable of, and Claude designs and builds the MCP server that exposes it. Once that bridge is committed and deployed, the loop closes: a feature request can go from conversation, to shipped code, to a live DNS entry, to a written and compiled release note — end-to-end, without an engineer manually running each step in between.
That loop isn't hypothetical. The logo at the top of this page was generated by Vivid, one of the bridges listed above, because Claude asked it for one. That's the real promise of agentic software: not that the model got smarter, but that it can now own the whole path from idea to running system.
Have a project like this one?
We'd be glad to talk through your goals, your constraints, and whether we're the right team for the work.