MCP for your agent
Connecting GeckoChat to Claude Code, Codex, OpenCode and Cursor
Updated September 22, 2026
MCP is the common way to hand an AI agent someone else's tools. Connect GeckoChat, and you talk to your agent in plain words — «show yesterday's orders», «put together a broadcast for everyone who didn't reach checkout» — while it calls the platform commands itself.
There is one address: https://api.geckochat.app/mcp, authorised by your key in the Authorization: Bearer … header.
Claude Code
claude mcp add --transport http geckochat https://api.geckochat.app/mcp --header "Authorization: Bearer gk_YOUR_KEY"
Add -s user to make the connection work in every project, not only in the current folder.
Codex
export GECKOCHAT_API_KEY=gk_YOUR_KEY codex mcp add geckochat --url https://api.geckochat.app/mcp --bearer-token-env-var GECKOCHAT_API_KEY
Codex reads the key from an environment variable, so declare it where Codex starts — in ~/.zshrc, for instance.
OpenCode
An opencode.json file, in the project folder or in ~/.config/opencode/:
{ "mcp": { "geckochat": { "type": "remote", "url": "https://api.geckochat.app/mcp", "headers": { "Authorization": "Bearer gk_YOUR_KEY" }, "enabled": true } } }
Cursor
The file ~/.cursor/mcp.json (or .cursor/mcp.json inside a project):
{ "mcpServers": { "geckochat": { "url": "https://api.geckochat.app/mcp", "headers": { "Authorization": "Bearer gk_YOUR_KEY" } } } }
Any other client
Anything that speaks remote MCP over HTTP will do: the address https://api.geckochat.app/mcp and the header Authorization: Bearer gk_YOUR_KEY. Nothing runs on your machine — the server is ours.
Checking it
Ask your agent to show your bots. It will call bot_list and answer with the list — that is the proof that the connection is live and the key was accepted.
MCP or the CLI
Both surfaces are the same commands; pick by the task.
| MCP | CLI | |
|---|---|---|
| Who decides what to call | the agent | you |
| Good for | a conversation: «work it out and do it» | a precise question with a large answer |
| Where it lives | inside your AI client | terminal, script, CI |
An MCP answer lands in the conversation with the model in full, so a hundred-record list is expensive there. The same list in a terminal collapses into three lines: --count-by, --sum and --table are computed on the server. If your agent has both, tell it plainly: «count it with geckochat in the terminal».