Pitaia MCP Server
Pitaia exposes a Model Context Protocol (MCP) server so you can connect your own AI agent — Claude, ChatGPT, or a custom agent — to your Pitaia store. Through MCP, an agent can read your store's analytics, manage products, save memories about your business, and draft charges for your approval.
This is the only supported public API surface of the Pitaia platform. The REST endpoints used by the Pitaia dashboard are internal and may change without notice.
Endpoint
| URL | https://api.pitaia.me/api/v1/mcp |
| Transport | MCP Streamable HTTP (stateless) |
| Method | POST only — GET returns 405 |
| Auth | Authorization: Bearer <agent token> — see Authentication |
Each request is handled by a fresh, stateless server instance: there are no MCP sessions to establish or resume, and server-initiated SSE streams are not used.
:::info Availability
The MCP server is rolling out gradually. If it is not yet enabled for your
environment, the endpoint returns 404.
:::
Quickstart
First create an agent token — the fastest way is the Connect an agent page at pitaia.me/dashboard/agent, which can also copy a ready-made MCP config for you. Then connect your client:
Claude Code
claude mcp add --transport http pitaia https://api.pitaia.me/api/v1/mcp \
--header "Authorization: Bearer YOUR_AGENT_TOKEN"
Claude Desktop / other MCP clients
Add to your MCP configuration:
{
"mcpServers": {
"pitaia": {
"url": "https://api.pitaia.me/api/v1/mcp",
"headers": {
"Authorization": "Bearer YOUR_AGENT_TOKEN"
}
}
}
}
Once connected, ask your agent something like "How did my store do this
month?" — it will call the analytics_overview
tool.
What the agent can do
The server exposes 16 tools, split into read scope (analytics,
product catalog, memories, store goal) and write scope (create/update
products, save memories, generate theme proposals, draft charges). Actions
with money or customer impact are human-gated: for example,
generate_charge only creates a draft — the checkout link is sent only after
you approve it in the Pitaia dashboard.
Every tool call is scoped to the store the token belongs to and recorded in the store's audit log.