MCP · Software for agents

An MCP server
your agents draw with.

Connect Claude, Cursor or any MCP client and call clearly_generate_svg / clearly_generate_image. Editable vector or raster art comes back — headless, no browser, billed against your AI pool. The same engine behind the Clearly studio, exposed as two tools.

Get an MCP tokenRead the docs
Fox mascot
Mountain landscape
Rocket launch
Botanical bloom
Geometric badge
Wave pattern

Every clearly_generate_svg call returns code like this — real, editable paths.

Two tools: SVG + rasterHeadless — no browserEditable SVG, not flattenedPer-generation cost returned
The call

Two tools. One token. No SDK.

The agent finds the tools in its list and calls them like any other. Generation runs server-side; you get editable markup or a hosted URL back — plus the exact cost.

tools/call
POST https://relay.clearly.sh/mcp
Authorization: Bearer ck_mcp_...

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "clearly_generate_svg",
    "arguments": {
      "prompt": "a fox in a wool scarf, friendly",
      "style": "flat-icon",
      "aspectRatio": "1:1"
    }
  }
}
result
// tool result
{
  "ok": true,
  "format": "svg",
  "svg": "<svg viewBox=\"0 0 512 512\">…editable paths…</svg>",
  "key": "u/abc/gen_8f21.svg",
  "costUsd": 0.04
}
The toolset

Vector when it must stay editable, raster when it must look photoreal

Pick the tool to match the job. Both are headless, both return a stable key + cost, both take an optional compositionId to drop the result on a canvas.

FeatureReturnsBest forKey args
clearly_generate_svgEditable SVG markupIcons, logos, stickers, cut filesprompt, style, aspectRatio, width/height
clearly_generate_imageHosted PNG + URLProduct shots, illustration, hero artprompt, model, aspectRatio, seed, width/height

Both tools live in the same MCP server at https://relay.clearly.sh/mcp. Omit compositionId to just receive the asset; include it to also place it.

Why MCP, not just an API

Built for the agents you already run

No new SDK, no rewrite, no flattened output — your agent gains a design tool the moment it connects.

Inside your agent loop

It’s an MCP server, so the tools appear natively in Claude Code, Cursor, Claude Desktop or your own client. The agent calls them mid-task — recall context, generate the asset, keep going — without leaving its loop.

Editable, not flattened

clearly_generate_svg returns real <svg> path code — your agent can post-process, diff, recolor, or hand it to a cutter. Not a PNG screenshot of a vector.

Metered, auditable spend

Every call returns costUsd and debits your AI pool — no separate per-seat fee for agents. Attribute spend per generation, per user, per job.

Agent makes, human refines

Pass a compositionId and the asset lands on a Clearly canvas, where a teammate can drag nodes and restyle paths. The agent drafts; a person finishes.

What agents build with it

From one-off icons to standing asset pipelines

Anywhere an agent needs to make a picture instead of describing one.

Design-system icons

Generate a whole icon set in one consistent style — pin style: 'flat-icon' across the fleet so 200 calls all match. Editable SVG drops straight into your component library.

Build-time asset pipelines

Wire generation into CI: an agent reads a manifest and emits the missing illustrations, og-images, or spot icons as part of the build — no designer in the critical path.

Chatbots that return art

A Slack, Discord or support bot that answers with an actual sticker or diagram, not a text description — generate on the fly and post the URL back.

Bulk on-brand catalogs

Spin up thousands of marketplace thumbnails, product badges, or category icons from a spreadsheet of prompts — same style, auditable cost per row.

Print & cut, headless

Generate vector art an agent can route to a cutter (Cricut, StickerMule, Roland) — kiss-cut stickers and decals produced end-to-end without opening an editor.

Creative coding agents

Give a coding agent a drawing primitive: it reasons about a layout, calls the tool, inspects the returned paths, and iterates — a real design loop, not prompt roulette.

Connect in a minute

Token, point, call

Step
01

Mint a token

Settings → Developers → Create MCP token. Pick rpc:write (generation needs it). Shown once, revocable, rate-limited.

Step
02

Point a client

Add the hosted endpoint to your MCP client — nothing to deploy. Claude Code, Cursor and Claude Desktop are one line each.

Step
03

Call the tools

Your agent now sees clearly_generate_svg / clearly_generate_image in its toolset and generates on demand, billed to your pool.

Claude Code
claude mcp add --transport http clearly https://relay.clearly.sh/mcp \
  --header "Authorization: Bearer ck_mcp_..."
FAQ

Agent generation, answered

01How is this different from a REST image API?+
It speaks MCP (Model Context Protocol), so it works inside the agents you already run — Claude Code, Cursor, Claude Desktop, or your own MCP client — with no SDK and no glue code. The agent discovers clearly_generate_svg / clearly_generate_image in its tool list and calls them like any other tool. (Prefer plain HTTP? The same engine is on the REST /api.)
02Is it really headless — no browser or canvas?+
Yes. The tools run entirely server-side: prompt in, asset out. clearly_generate_svg returns editable SVG markup; clearly_generate_image returns a hosted PNG URL. Both also return a stable key and the exact cost. Pass an optional compositionId to also drop the result onto one of your canvases; omit it to just get the asset back.
03What exactly comes back?+
A JSON result: { ok, format, svg | imageUrl, key, costUsd }. For vectors you get real <svg> path code your agent can post-process, diff, or hand to a cutter; for raster you get a files.clearly.sh URL. The cost is returned on every call so spend is auditable per generation.
04How do I connect an agent?+
Mint an MCP token in Settings → Developers (scopes: rpc:read / rpc:write — generation needs rpc:write), then point a client at https://relay.clearly.sh/mcp. Claude Code: `claude mcp add --transport http clearly …`. Cursor / Claude Desktop: add the server to mcp.json. Nothing to deploy — the server is hosted.
05How is it billed?+
Each generation debits the AI pool of the token owner (your Clearly plan), and the per-call costUsd is returned so you can attribute it. There is no separate per-seat charge for agent calls — agents spend the same metered pool your team does. Top-ups cover overflow.
06Can I control the art style?+
Yes — clearly_generate_svg takes a style (vector-illustration, flat-icon, line-art, isometric, sticker, engraving) so a fleet of agent calls stays visually consistent, which is what makes it usable for design systems and bulk catalogs. clearly_generate_image takes a model + seed for reproducibility.
07Do the results stay editable?+
The SVG ones do — they are vector path code, not a flattened PNG of a vector. If you pass a compositionId, the asset also lands on a Clearly canvas where you (or a teammate) can drag nodes and recolor paths. That round-trip — agent generates, human refines — is the whole point.
08Which clients are supported?+
Any MCP client. We test Claude Code, Cursor and Claude Desktop; the Claude Code plugin also ships skills. Your own code can speak the MCP JSON-RPC directly against the hosted endpoint with a Bearer token.
Explore the platform

Part of the Clearly agent stack

Software for agents

The pillar — how agents generate, perceive, and drive a Clearly canvas across MCP, API, and CLI.

MCP docs

The full reference — every tool, scope, and agent-identity option for the hosted server.

REST API

Prefer plain HTTP? The same generation engine behind one REST endpoint, with per-image cost.

Beehaven CLI

Drive generations and your agent cloud from the terminal — the CLI over the same surface.

Give your agent a design tool

Mint an MCP token, point your client at the hosted server, and start generating editable SVG and raster art in your agent loop — metered against your AI pool.

Get an MCP tokenRead the docs ⟶