System prompts your team and your agents manage together.
Every change, human or agent, is a version with a commit message and a one-click rollback. Your app reads it over REST or the Node SDK; your agent reads and edits it through MCP.
Move each prompt id across in a few minutes, keep variables and versions, and stop depending on one vendor's dashboard. The migration guide walks through the three steps.
Write, version, publish and fetch system prompts. That is the whole product, and it is done properly.
Add one MCP endpoint and your agent can list, read, edit and publish prompts from the terminal or a chat — no API keys needed if you use OAuth.
Write your prompt as markdown: every heading is a section. No drag handles, no fixed templates — just a clean textarea that converts headings to sections on save.
Write {{ customer }} anywhere. The API lists the variables a prompt needs and the SDK fills them in at runtime, strictly if you want.
Paste or drop a markdown file. Headings become sections. Prompts you already have are in the editor in a few seconds.
Versions are content-addressed hashes, like git. Compare any two, write a commit message, and see who changed what.
Production points at one version. Publish a new one when it is ready, roll back in one click when it is not. Apps pick it up on the next request.
Ask for ?version=latest in staging and production by default everywhere else. Same prompt id, no separate copies to keep in sync.
GET /api/v2/prompts/:id returns the assembled text, sections, tools and variables. Your app reads over REST or the Node SDK; your agent writes through the same endpoint via MCP.
npm install superprompts. Cached reads, variable compilation, typed responses, zero dependencies. Every other language reads the same prompt over REST.
Run any version against OpenAI, Anthropic, Google, Mistral or xAI side by side before you publish. Your provider keys, your bill.
Optional anti-injection instructions appended to served prompts. Blocks the usual "reveal your system prompt" attempts. Turn it off per request.
One project per app, one API key per project. Invite teammates so the person who owns the words can change them without a deploy.
One response has everything your call to the model needs. No client library required, though the Node SDK saves you the caching and variable code.
Protect your AI applications from malicious attacks. Our intelligent guardrails system automatically detects and blocks prompt injections, prevents system prompt leaks, and secures sensitive information.
Detects and neutralizes attempts to override system instructions or manipulate model behavior.
Prevents attackers from extracting your proprietary system prompts and instructions.
Automatically identifies and blocks attempts to access environment variables or keys.
SuperPrompts serves text and tool definitions; it never sits between your app and the model. The playground calls providers with keys you add.
Chat Completions and Responses APIs, with your own key
Messages API, with your own key
Google AI Studio API, with your own key
La Plateforme API, with your own key
xAI API, with your own key
Open the same version in several chat windows, one per provider, fill in the variables, and read the answers side by side before you publish.
Four steps, and the last three repeat for the life of the prompt.
Paste the prompt you have or start from sections. Mark per-request values as {{ variables }}.
Save creates a version. Publish points production at it. History keeps every version with a diff.
Replace the hardcoded string with one call: REST from any language, the Node SDK, or your agent over MCP.
Try changes in the playground against real models, publish when it reads right, roll back when it does not.
A system prompt is several concerns stacked together: role, rules, examples, output format. Sections keep them apart so a diff shows what actually changed, and so a teammate can edit the tone without touching the rules.
Read, write and publish prompts over HTTP with a project key.
Cached reads, variables, typed responses, no dependencies. Every other language uses the same REST call.
One paid plan, priced for a small team, not for a platform budget. Start free, no credit card.
Enough to ship a real side project
For teams and production applications
All plans include:
Sign in, paste the prompt you have, publish it, swap the hardcoded string for one call. The free plan covers a real project.
No credit card, no sales call
Honest side-by-side comparisons. We'll tell you where SuperPrompts wins and where it doesn't.
Versus the LangChain observability platform with prompt management as one feature.
Versus the open-source observability + prompts platform (Apache-2.0, self-hostable).
Versus the full prompt-ops suite with AB testing, workflows, and tracing.
Versus the git-for-prompts tool: repos and branches, with agent editing as an afterthought.