Prabloé, from inside your AI.
Prabloé runs one MCP server at https://www.prabloe.com/api/mcp. Point Claude, Claude Code or Cursor at it and their tool list picks up everything below — no key required for most of it.
Connect a client
These three are the ones that actually work today — a server on a private network or one that expects a sign-in flow would be worse to list than to leave out.
- 1.In Claude, open Settings → Customize → Connectors, then click Add custom connector.
- 2.Name it Prabloé and paste the URL below.
- 3.Save. Claude reaches this server from its own cloud, so nothing needs to run on your machine.
https://www.prabloe.com/api/mcp
Ask it things like "ask Prabloé to explain X", "compress this prompt into a brief for my agents", or "search the live web for Y" — those work signed out. Generating an image or video, or using memory, needs a signed-in Prabloé account, which this connection does not carry.
What it can call
Every tool the endpoint registers, read straight from the server. Four work for anyone; the rest need a signed-in Prabloé account, which a client connected this way does not have.
| Tool | What it does | Needs sign-in |
|---|---|---|
| ask_prabloe | Ask Prabloé a single question and get one reply (non-streaming). Prabloé explains step by step in plain language — good for explanations, worked examples, and reasoning through a problem. | No |
| compress_prompt | Turn a raw, possibly messy prompt into a compact, high-information-density brief (goal, constraints, artifacts, a self-verifiable success check, and a proposed multi-agent fan-out with a self-contained sub-brief per agent). Useful before handing a task to a set of specialist agents. | No |
| remember | Save one short fact, preference, rule or dictionary spelling to the user's Prabloé memory. It is private to them and is read back by Prabloé chat, by Prabloé Code agent runs and by voice dictation cleanup. Requires a signed-in caller. | Yes |
| recall | Read back what the user has asked Prabloé to remember — everything, or the entries matching a search term. Private to the signed-in caller. | Yes |
| capabilities | List what this caller can actually do right now: whether they are signed in, their credit balance, whether image/video creation is switched on for them, which generation models are available, and which engines are configured. Call this FIRST when planning work, so a plan is built from what is really available rather than a guess. | No |
| generate_image | Create an image from a text prompt with Prabloé Create, Prabloé's own image generation — or edit finished pictures (referenceJobIds), or upscale / cut out the background of one (tool). Returns a job id immediately; call `job_status` with that id and `waitSeconds` for the result. Never returns image bytes. Requires a signed-in account: this creates a durable record and reserves credits. | Yes |
| generate_video | Create a short video from a text prompt with Prabloé Create, Prabloé's own video generation — optionally animating a finished picture (startFrameJobId), or run a clip tool (caption, reframe, video-upscale) on a finished video (sourceJobId). Genuinely asynchronous — returns a job id immediately; call `job_status` with that id and `waitSeconds`. Never returns video bytes. Requires a signed-in account: this creates a durable record and reserves credits. | Yes |
| job_status | Check one Prabloé Create generation job, optionally waiting up to 35s (waitSeconds) for it to finish. Answers { status, kind, resultUrl, resultUrls, error, next }. `resultUrl` is a SHORT-LIVED SIGNED URL to the finished asset, minted only after ownership is checked — never the raw bytes, and never a vendor URL. `next` says what to do now. A job belonging to someone else is reported as not found. Requires a signed-in account. | Yes |
| research | Search the live web and get back a synthesised answer with the source URLs it cited. Use this instead of answering from memory for anything that moves fast — model names, prices, API shapes, whether a thing already exists. | No |