PromptOT MCP Server

PromptOT를 통해 LLM 프롬프트를 관리, 버전 관리 및 게시하세요. 블록으로 구조화된 프롬프트를 만들고, 변수를 추가하고, 초안을 저장하고, 버전을 게시하고, API를 통해 전달할 수 있습니다 — 모두 채팅에서 가능합니다.

promptot.com에서 보기

문서

Back to Documentation

MCP Server

PromptOT MCP Server.

Use PromptOT directly from Claude Desktop, Cursor, Codex CLI, ChatGPT, claude.ai, and any other Model Context Protocol-compatible AI tool. 23 MCP tools cover the full prompt lifecycle: list, create, edit, version, publish, and inspect prompts from inside your AI assistant.

What is MCP?

The Model Context Protocol is an open standard for connecting AI assistants to external tools and data sources. PromptOT's MCP server exposes our prompt management API as a set of typed tools that any MCP-compatible client can call. Once installed, the AI model can list your prompts, edit blocks, save draft versions, publish to production, and run test cases — all through normal conversation.

The server ships in two forms: a local stdio binary distributed via npm (npx @prompt-ot/mcp) for desktop clients, and a hosted HTTP transport at mcp.promptot.com for browser-based clients that authenticate via OAuth.

Quick start

Generate an MCP key

From your dashboard, open any project's API Keys page and click "Generate MCP Key" in the MCP Server card. Optionally enable AI features if you want playground/evals/AI rewrites.

Paste the install snippet

Click your client (Claude Desktop, Cursor, Codex CLI, etc.) and copy the snippet. Paste it into your client's MCP config file at the path shown.

Restart and ask your AI

Restart your AI client. The PromptOT tools become available automatically. Try "List my PromptOT prompts" or "Create a new customer support prompt".

Generate MCP key →

Reference

Install

Step-by-step install snippets for all 8 supported clients: Claude Desktop, Cursor, Codex CLI, ChatGPT, claude.ai, Windsurf, Zed, and any generic stdio client.

Tools reference

Full catalog of all 23 MCP tools with input parameters, required scopes, and one-line descriptions. Grouped by area: prompts, blocks, variables, versions, test cases.

Scopes

The permission vocabulary that gates MCP API calls. Read scopes vs write scopes vs AI scopes, plus the default MCP key preset and the AI features opt-in.

OAuth flow

How browser-based clients (claude.ai, ChatGPT) authenticate against the hosted MCP server at mcp.promptot.com via OAuth 2.1 + PKCE. Discovery endpoints, consent flow, token exchange.

Troubleshooting

Common errors and how to fix them: 401 unauthorized, 403 insufficient scope, AI credit exhausted, rate limits, missing tools in Claude Desktop, truncated outputs.

See also: REST API reference for the underlying HTTP endpoints.