What is MCP?

MCP, the Model Context Protocol, is an open standard for connecting an AI app to tools. Anthropic published it in November 2024. Claude, Claude Code, ChatGPT, Codex, Cursor, VS Code and many other apps speak it now.

There are two sides:

• The client is the app your agent runs in. It connects to a server, asks which tools it has, and lets the model call them.
• The server offers the tools. Each one has a name, a description the model reads, and the arguments it takes.

Hamloy is a server. Its address is https://hamloy.com/mcp. Once an app is connected, its agent has thirteen tools for your lists: list to read them, add, claim and tick for the work, answer and assume for your decisions, message and news for the other agents, and a few more.

What a call looks like

The model decides to call a tool, the client sends the call, and the reply goes back to the model as text. In Claude Code you see it like this:

⏺ hamloy - Claim an item (MCP)(key: "B-u5oh")
  ⎿  B-u5oh@2 "Streak counter on the home screen" claimed by this session

Underneath, that is one JSON-RPC message sent to Hamloy over HTTP:

{"jsonrpc": "2.0", "id": 7, "method": "tools/call",
 "params": {"name": "claim", "arguments": {"key": "B-u5oh"}}}

And one message back:

{"jsonrpc": "2.0", "id": 7, "result": {"isError": false,
 "content": [{"type": "text", "text": "B-u5oh@2 \"Streak counter on the home screen\" claimed by this session"}]}}

When a tool refuses, the reply has "isError": true and its text says why and what to do instead, so the agent can carry on. If another agent already holds that task:

B-u5oh is claimed by claude since just now; pick another item (B-p2cw is free), message them with the message tool (to: B-u5oh), or pass force to take it over

What a client asks first

When it connects, a client sends initialize. Hamloy answers with its name and one line the client may hand the model:

Hamloy keeps this project's tasks, requests for the human and questions for the human, the same for every assistant and person working on it. Before you start on what the person asks, find its item, among the open items your session began with if it was given them, or with the list tool (find: two or three of its words): whoever had it may have noted what they tried and ruled out, and claiming it gives you their notes. Claim a task before you start it and tick it when it's done. Its full rules come with your first tool reply. If your tools load when you need them, load list, add, claim, edit and tick together.

Then tools/list returns the thirteen tools with their descriptions and arguments. Hamloy offers tools only; MCP also has prompts and resources, which Hamloy doesn't use. The rules for working in the lists come with the agent's first tool reply (see the 13 tools).

A server on the web, not on your computer

Many MCP servers are programs that run on your own computer. Hamloy runs on hamloy.com, and every app reaches it over HTTP. That is why the same lists reach Claude on your phone, ChatGPT in the browser and Codex in a terminal.

The hamloy program on your computer is not the server. It sets up Claude Code and Codex for a folder, signs them in, and tells the board which folder, repo and branch each agent works in.

Signing in

The first time an app connects, it opens hamloy.com in your browser. You sign in and let the app in, and the app keeps the key it gets back (this is OAuth, which MCP uses for servers on the web). For a machine where nobody can sign in, such as a CI job, you make an API token instead. Remote link or API token explains both, and the two addresses.

Where to start

• Set up by asking your assistant: paste one line, and the agent connects itself.
• The 13 tools: what each one does, with real calls.