Managed Prompt Registry

Your prompts deserve
 

AI Prompt Bucket is the managed registry that gives your team versioned templates, safe deployments, health scoring, and a REST API + MCP server — so your prompts stop living in random strings scattered across your codebase.

Start Free See How It Works

Free tier includes 25 prompts. No credit card required.

Works with every MCP client and REST API

Claude Cursor Windsurf Any MCP Client REST API

Your prompts are flying blind

When a prompt breaks in production, you find out from angry users — not from your tools. There's no version history, no rollback, no way to know which prompts are stale or who changed what last Tuesday.

Prompts buried in code

Scattered across files, repos, and config. Nobody knows what's live in production.

No version history

One bad edit and the old prompt is gone. No diffs, no rollback, no audit trail.

Updating means redeploying

Changing a prompt requires a code change, a PR, a build, and a deploy. For a string.

your_app.py — before APB 
# Which version is this? Who changed it?
# Is this the same in staging?
# Nobody knows.

ONBOARDING_PROMPT = """
You are a helpful onboarding assistant.
Welcome  to the  plan.
Help them set up their account.
"""

# Hardcoded. No versioning. No rollback.
# Change this and hope for the best.
response = llm(ONBOARDING_PROMPT.format(
    name=user.name,
    plan=user.plan
))

How it works

From blank canvas to production-ready prompt in four steps.

1

Create

Write your prompt template with Jinja2, f-string, or Mustache variables. Define a schema for the inputs you expect.

2

Test

Use the Playground to render with test variables. Compare versions side-by-side with unified diffs.

3

Deploy

Assign labels like production or staging to any version. Roll back instantly if needed.

4

Integrate

Fetch rendered prompts via REST API or MCP. Your app calls one endpoint — we handle the rest.

Everything your prompts need

From a simple dashboard to a full REST API and MCP server.

Immutable Versioning

Every save creates a new version. Compare diffs, rollback instantly, and see exactly who changed what and when.

Flexible Labels

Point labels like production and staging at any version. Create custom labels for any workflow. One prompt, many pointers.

Template Rendering

Jinja2, f-string, and Mustache templates with sandboxed rendering. Pass variables via API and get back ready-to-use text.

Health Scoring

Every prompt gets an A–F grade across completeness, maintenance, deployment, and coverage. Spot problems before users do.

Full REST API

Complete CRUD API with Swagger docs, team-scoped API keys, and render-on-fetch. One endpoint, always the latest version.

MCP Server

Your AI tools manage prompts directly. Claude, Cursor, Windsurf — any MCP client connects instantly. No plugins needed.

Claude Desktop — MCP 
You: Create a customer support prompt for refund
     requests. Use Jinja2. Deploy to staging.

Claude: I'll create that prompt for you.

  • create_prompt    slug: "refund-handler"   ✓
  • create_version   v1 with Jinja2 template  ✓
  • render_prompt    test render with sample   ✓
  • promote_version  v1 → staging            ✓
  • get_prompt_health score: B (82/100)       ✓

Done. "refund-handler" is live on staging.
The completeness score flagged a missing output
format — want me to add structured JSON
output to bring it to an A?

Your AI tools manage your prompts

Most prompt tools give you a dashboard and maybe an API. AI Prompt Bucket ships a full MCP server — so Claude, Cursor, and Windsurf can create, version, deploy, and analyze your prompts without leaving your editor.

Efficient MCP tools covering every operation

Prompt CRUD, version management, label assignment, health scoring, analysis, and team admin.

Enterprise-grade auth

OAuth 2.1 + PKCE for interactive clients. API key fallback for scripts and CI/CD.

Works with your tools today

Claude Desktop, Claude Code, Cursor, Windsurf — any MCP-compatible client connects instantly.

"Can't I just use Git / a database table?"

You could. Here's why teams don't.

Git isn't a prompt registry

Versions files, not runtime configs.

Git versions files. Prompts need runtime fetching, label-based promotion, template rendering, and health scoring. A git commit can't do that.

Versions, deploys, renders, scores, and serves — no redeploy needed.

A database table isn't a product

2–4 weeks of engineering, minimum.

You'd build versioning, rollback, label-based promotion, rendering, an API, key management, and access controls. That's a product, not a table.

Sign up, create a prompt, call the API. Done in 5 minutes.

Build vs. buy

12 features you'd have to maintain.

Immutable versioning, label-based promotion, rollback, template rendering, sandboxed execution, API + auth, version diffing, health scoring, duplicate detection, workspace export, usage analytics, MCP integration.

Every team that builds this in-house ends up with a worse version.

Simple, generous pricing

Start free. Scale as your prompt library grows.

Monthly
Annual Save 17%
Free
$0 / forever

For solo devs and experiments.

  • 25 prompts
  • 50 versions per prompt
  • 3 labels
  • 1 team member
  • 1,000 API calls / mo
  • REST API + MCP server
  • Health scoring
  • Playground
Get Started
Most Popular
Pro
$19 / month

For small teams shipping AI products.

  • 200 prompts
  • Unlimited versions
  • Unlimited labels
  • 5 team members
  • 25,000 API calls / mo
  • Automated analysis
  • Workspace export
  • Prompt graph
Start Free Trial
Team
$39 / month

For growing teams with many prompts.

  • 1,000 prompts
  • 10 team members
  • 100,000 API calls / mo
  • Everything in Pro
  • apb scan CLI import
  • Version comparison diffs
  • Priority support
Start Free Trial
Business
$99 / month

Unlimited prompt management for organizations.

  • Unlimited prompts
  • Unlimited everything
  • Everything in Team
  • Audit log export
  • Dedicated support
  • Custom integrations
Start Free Trial

Enterprise?

We’re built for indie developers and small teams. But if you need custom volumes, SSO/SAML, or dedicated support — let’s talk.

Frequently asked questions

How does versioning work?

Every time you save a prompt, a new immutable version is created automatically. You can compare any two versions with a unified diff, rollback to any previous version in one click, and see exactly who changed what and when. Versions are never overwritten.

What are labels?

Labels are flexible pointers like latest, staging, and production that point to a specific prompt version — so you can test v5 under staging while v3 is still live under production. Custom labels are supported on paid plans.

How does the MCP server work?

AI Prompt Bucket exposes a full MCP server that any compatible client (Claude Desktop, Claude Code, Cursor, Windsurf) can connect to. Your AI tools gain the ability to create, version, deploy, analyze, and manage prompts directly — no context-switching needed. Authentication uses OAuth 2.1 + PKCE.

What template formats are supported?

Jinja2 (sandboxed), Python f-strings, and Mustache. Templates are rendered server-side when you fetch a prompt via API, so your app gets back ready-to-use text. Variable schemas let you define expected inputs with types and defaults.

What is health scoring?

Every prompt gets an A–F letter grade based on four factors: completeness (description, schema, tags), maintenance (staleness, recent edits), deployment (does it have a production label?), and coverage (variable usage). It helps you spot problems before they reach users.

Is the free tier really free forever?

Yes. The free tier includes 25 prompts, 50 versions per prompt, 3 labels, the full REST API, MCP server, health scoring, and the Playground. No credit card required, no trial expiration. Upgrade when you need more prompts, team members, or advanced features.

Stop scattering prompts. Start managing them.

25 prompts free, forever. Your team will wonder how they shipped AI products without this.

Start Free