Services / MCP development
MCP development, by a team that ships MCP products.
The Model Context Protocol is how AI assistants like Claude and ChatGPT connect to real systems — and it’s quickly becoming the integration layer serious teams standardize on. We design and build the servers and connectors that make it safe and useful.
Proof: Webtold and SchoolWare — both ours, both MCP-native. This site is maintained over MCP.
What we build
Servers, connectors, and the unglamorous parts around them — auth, evals, hosting. We usually recommend starting with one system and growing from there.
Custom MCP servers
Expose your internal APIs, databases, and tools to AI assistants — with schemas designed so agents use them correctly the first time.
Product connectors
Let your customers use your SaaS from inside Claude or ChatGPT. A good connector is becoming a distribution channel, not a feature.
Auth, tenancy & audit
OAuth flows, per-user and per-tenant permissions, rate limits, and audit trails — an assistant should never out-privilege its user.
Tool design & evals
Naming, descriptions, and guardrails tested against real assistant behavior, so tools get called correctly — and refused correctly.
Hosting & operations
Deployment, monitoring, versioning, and upgrades as the spec evolves. MCP servers are production services and we run them like it.
Audits & rescues
Already have an MCP server that misbehaves? We review tool design, security, and reliability — and fix what we find.
Is MCP right for your problem?
When MCP is the right call
You want an assistant to act on live systems, not just chat — pull real numbers, create records, run workflows. Your customers already live in Claude or ChatGPT and keep asking for a connection. Or your internal ops span a dozen tools and you want one conversational surface across all of them.
When it isn’t
If you need a simple FAQ chatbot, MCP is overkill — a retrieval setup is cheaper and faster. And if your data can’t safely leave its current boundaries yet, we’ll tell you to fix that first. We’d rather lose a project than ship you the wrong thing.
An engagement like this
A B2B SaaS exposes its reporting API as an MCP connector. Its customers stop exporting CSVs and start asking Claude for the numbers directly — and “how do I build this report” support tickets drop off a cliff. That’s the shape of the work: pick the highest-friction workflow, wire it, measure it.
Questions we hear a lot
What is an MCP server, in plain English?
MCP (Model Context Protocol) is an open standard that lets AI assistants like Claude and ChatGPT connect to real systems. An MCP server is the adapter: it exposes your APIs, databases, or product as safe, well-described tools an assistant can actually use — read a record, create an order, publish a page — with permissions enforced on every call.
How long does it take to build an MCP server?
A focused server exposing one system typically takes 2–4 weeks including auth, testing, and documentation. A customer-facing connector for a SaaS product, with OAuth, per-tenant permissions, and hardening, usually runs 4–8 weeks. You get a fixed scope and date before we start.
Does MCP work with both Claude and ChatGPT?
Yes. MCP is an open standard originally created by Anthropic and now adopted across major AI clients. We build to the spec, so one well-built server works across MCP-aware clients rather than needing a rewrite per assistant.
How do you handle security and permissions?
Every tool call is authenticated and authorized — OAuth or API keys, per-user and per-tenant scopes, rate limits, audit logs, and careful tool design so an assistant can never do more than the signed-in user could. We treat MCP servers as production services, not demos.
What does MCP development cost?
We quote a fixed price after a free, plain-English proposal — you tell us the systems and the outcomes, we tell you the scope, the price, and the date. No open-ended retainers to get to version one.
Tell us what your assistant should be able to do
We’ll come back with the scope, the price, and the date — in plain English, free.