Services · Intelligence
The Model Context Protocol is how AI assistants like Claude and ChatGPT call outside tools. An MCP server lets them read your facts and take the actions you allow, from your data, under your rules. We build them, list them, and run them.
- A company whose customers already ask AI assistants about its products, and get answers from somewhere else.
- A software product that wants to be usable from inside Claude, ChatGPT, Cursor or VS Code.
- A team that wants its own staff assistants to reach internal systems through one controlled door.
What we do
The work, in plain words.
- 01
Design the tools
What an assistant may read, what it may do, and what it must never see. Each tool gets a title, a clear description, input and output schemas, and annotations that say whether it is read-only or acts on the world.
- 02
Speak both eras of the protocol
Clients move at different speeds. We serve the current specification and the earlier handshake on one endpoint, so new and older assistants both connect.
- 03
Guard it like a public API
Origin checks, body limits, rate limits per caller and per day, duplicate protection on any tool that writes, and logs that record which tool ran and for which app, never what was asked.
- 04
List it where assistants look
A server card and discovery file on your domain, the official MCP Registry, and the directories assistants browse, with a human page that explains each tool.
- 05
Test and run it
Unit and integration tests against the official client, live checks after every deploy, and the server kept current as the specification moves.
How it starts
Three steps, each in writing.
- Step 1
A conversation
Who should reach you through an assistant, and what they should be able to do.
- Step 2
A tool list
The tools, their inputs and outputs, and what each may never return, written down before we code.
- Step 3
Build, list, run
The server on your domain, its listings, and the monthly run.
Proof
Running today, not a promise.
bluecoastwater.com/mcp
Our own public MCP server: facts about the firm from this site, and a way to write to the founder, both protocol eras, with the same contact tool offered through WebMCP on the contact page.
Blue Coast Water for AI assistants →samplyn.com/mcp
Our own product's public MCP server: product facts and a demo request, both protocol eras, listed in the official MCP Registry.
Samplyn for AI assistants ↗A catalog MCP behind a fashion client's brand sites
A read-only catalog server that each brand site proxies at its own /mcp, so assistants can browse the line without ever seeing a price.
Agent search, end to end
The MCP is one piece of making a business findable by AI. The rest is on the AI search page.
AI search optimization →
What we do not do
Where we draw the line.
- We do not expose more than the tool list says. Every output is checked against its schema and the copy rules in tests.
- We do not build an MCP that writes without limits. Any tool that acts is rate-limited, deduplicated and logged.
- We do not promise a listing will be approved by a directory; we prepare what each asks for and submit it.
Questions
What people ask first.
What is an MCP server, in plain words?
A small web service that tells an AI assistant which tools it may use, and answers when the assistant calls them. Add its address in Claude or ChatGPT and the assistant can use your tools.
Which assistants can use it?
Claude, ChatGPT in developer mode, Cursor, VS Code and other MCP clients. We test with the official client in both protocol eras.
Can it write data, or only read?
Both, by design. Read tools are marked read-only; a tool that acts, like sending a request to your team, is marked as such and guarded with limits.
Does it need users to sign in?
A public server with public facts does not. A server over private data needs sign-in and per-user permissions, which we scope with you.
Where does it run?
On your own domain, usually as a small serverless function beside your site, in your cloud account.
What does it cost?
The founder quotes after the first conversation, in writing. There is no price list on the site.
No prices on this site: the founder quotes in writing after the first conversation. How engagements work.
Tell me what you need.
Philip Radvan, founder. Every message lands with me, and it's me on the call.