# Keep a prompt library for Claude, ChatGPT and Cursor

Each prompt, every version, the model it’s for, how it scored and who changed it. Your app reads the live version with a key that can’t write.

## An example

**You:** Save v4 of the support-reply prompt: added "never promise refunds" and moved the tone guidance first. Scored 8.6 on our rubric vs 8.1 for v3.

**ADoor:** Added 1 entry to Prompt versions

**Prompt versions**

| Prompt | Version | Model | Score |
| --- | --- | --- | --- |
| support-reply | v4 | gpt-5 | 8.6 |
| support-reply | v3 | gpt-5 | 8.1 |

## What you can ask later

- “What’s the current version of support-reply, and what changed from v3?”
- “Which prompts haven’t been evaluated on the new model?”
- “Who changed the summarizer prompt last?”
- “Show the best-scoring version of each prompt.”

## Who can see what

- **Prompt owners** add versions and mark one as live.
- **Your app** reads live versions with an API key that can’t write.
- **Agents** can propose a new version as a draft, which a person approves.
- The history keeps recent changes with the before and after.

## How it compares with LangSmith or PromptLayer

For playgrounds, per-call tracing and evals on every prompt change, LangSmith and PromptLayer are the tools to use.

As a simple, access-controlled source of truth, ADoor works: versions with scores, an app key that only reads, and a history of who changed what.

## How to create a prompt library in ChatGPT and Claude

1. Connect ADoor to ChatGPT and to Claude, choosing the same space.
2. In either one, ask: “Set up a prompt library: name, prompt, model, version, score and notes.”
3. Save a prompt from ChatGPT, and use it from Claude or Cursor. Every edit is a new version in the history.

## How to start

1. Sign in to ADoor, then add it to an assistant like ChatGPT, Claude, Grok or Cursor through MCP.
2. Tell it: “Set up a prompt library: Prompts (name, purpose, owner) and Prompt versions (prompt, version, text, model, score, live, notes).”
3. Create a read-only key for your app, and let agents add draft versions that a person marks live.

[Get your account](https://app.adoor.halmoslabs.com/signup)

## Questions people ask

### Can my app fetch the live prompt?

Yes, through the HTTP API with a read-only key.

### Can I roll back?

Mark an older version as live. The history keeps recent changes too.

### Do I have to pay?

Not on the free plan, which holds 3 tables and 100 rows. Pro is $10 a month per workspace, with 1 GB of storage and no per-GB charges.

### What happens if I remove someone’s access?

Their next request is refused. What they already saved stays, and the history shows who made each change.

## Related

- [Track eval results over time](/use/eval-results/)
- [Share agent traces across a team](/use/agent-traces/)
- [Keep feature flags and config agents can read](/use/feature-flags/)
- [ADoor vs Supabase](/compare/adoor-vs-supabase/)
- [ADoor vs Notion](/compare/adoor-vs-notion/)
- [All use cases](/use/)

Updated October 20, 2026.