MCP Server

PromptOT, in your
AI assistant.

23 Model Context Protocol tools that let Claude Desktop, Cursor, ChatGPT, and any MCP-compatible assistant manage your prompts, blocks, versions, and test cases — without leaving the chat.

How it works

Three steps to PromptOT in your AI tool.

No SDKs, no glue code. Generate a key, paste a snippet, and start asking your assistant to manage your prompts.

Step 1

Generate an MCP key

Open your project's API keys page and create a key with the MCP scope. Pick read, write, or publish — defaults are conservative.

Step 2

Paste the install snippet

Drop the generated config into Claude Desktop, Cursor, Codex CLI, Windsurf or Zed. Or sign in via OAuth from claude.ai and ChatGPT.

Step 3

Ask your AI to manage prompts

Say things like "list my prompts", "create a new draft for the support bot", or "publish v3". Your assistant uses the 23 MCP tools automatically.

Tool catalog

23 tools, grouped by what they touch.

The PromptOT MCP server exposes the same primitives the dashboard does — so anything you can do as a human, your AI assistant can do too.

Prompts6
  • list_prompts
  • get_prompt
  • get_compiled_prompt
  • create_prompt
  • update_prompt
  • delete_prompt
Blocks5
  • list_blocks
  • create_block
  • update_block
  • delete_block
  • reorder_blocks
Variables3
  • list_variables
  • upsert_variable
  • delete_variable
Versions5
  • list_versions
  • save_draft_version
  • publish_version
  • rollback_to_version
  • diff_versions
Test cases4
  • list_test_cases
  • create_test_case
  • update_test_case
  • delete_test_case
Security & control

Built so you can hand the keys to an AI safely.

We assume the assistant will sometimes do unexpected things. Every PromptOT MCP key has guardrails so it can't accidentally publish to production or run up your bill.

Scoped API keys

Choose read, write, or publish scopes per key. The MCP server only sees what you grant.

Development by default

New keys default to the development environment and cannot publish to production unless you explicitly opt in.

Full audit log

Every MCP-originated mutation is recorded with the client name, scope, and user — visible at /audit-log.

One-click revoke

Pull access at any time from the dashboard. Revocation is immediate across every connected client.

AI features off by default

Playground, evaluations, and AI rewrites are disabled per key until you turn them on, so MCP traffic can't burn through credits.

MCP works on every plan, including free.

Connect, list, edit, version, and publish prompts on the free tier without limits on MCP traffic. Upgrade to Pro to enable AI features — playground runs, evaluations, and AI rewrites — for your MCP key.

See pricing
Questions

Common questions.

The Model Context Protocol is an open standard from Anthropic that lets AI assistants talk to external tools and data sources. PromptOT ships an MCP server so any compatible assistant can manage your prompts.
Claude Desktop, Cursor, Codex CLI, Windsurf, Zed, claude.ai, ChatGPT, and any generic stdio MCP client. Each has a dedicated install page under /mcp/{client}.
No. For Claude Desktop, Cursor, Codex CLI, Windsurf, and Zed you copy a JSON or TOML snippet into a config file and restart the editor. For claude.ai and ChatGPT you sign in via OAuth — no config file at all.
Yes — the MCP server is a thin proxy to the same authenticated API your dashboard uses. Calls go over TLS and are scoped to the projects your key can access.
Open the API keys page in your project, click the key, and hit revoke. The MCP server rejects the next request immediately and your AI assistant will surface the error.
MCP itself works on every plan, including the free tier. AI features (playground, evals, AI rewrites) require the Pro plan and must be explicitly enabled per key — they're off by default.

Try PromptOT in your AI assistant in under a minute.

Generate a scoped MCP key, paste the snippet into your tool of choice, and start asking.