MCP server
Connect Claude, Cursor, and other MCP clients to the VectorFast Render remote MCP server, with setup per client, the seven tools and their annotations, and limits.
Short answer: the MCP endpoint is https://vectorfast.ai/mcp (Streamable HTTP, JSON responses, no session). Send your API key as Authorization: Bearer vf_live_…. The server exposes seven tools: six read-only tools and render.
VectorFast Render is not listed in any connector directory yet; add it by URL as a custom server.
Client setup
| Client | How | Status |
|---|---|---|
| Claude Code | claude mcp add --transport http vectorfast-render https://vectorfast.ai/mcp --header "Authorization: Bearer $VECTORFAST_API_KEY" | Works with an API key |
| Claude (web and Desktop) | Customize > Connectors > Add custom connector, No sign-in, Request headers authorization: Bearer vf_live_… | Request headers are an Anthropic beta for a limited set of organizations |
| Cursor | .cursor/mcp.json entry with url and an Authorization header | Configured per Cursor's docs; not yet tested end to end |
| ChatGPT | Plugins > Add custom MCP server | Not yet: ChatGPT needs OAuth or no auth and cannot send API keys |
Step-by-step pages with sources: Claude, Cursor, ChatGPT.
Tools
The order below is the fixed order of tools/list. Annotations are hints for clients: read-only tools can run without a confirmation prompt in clients that support it.
| Tool | Title | Read-only | Destructive | Idempotent | Open world | Description |
|---|---|---|---|---|---|---|
list_templates | List templates | yes | no | — | no | Lists the official templates with id, name, summary, and sizes. |
get_template | Get template details | yes | no | — | no | One template's parameter JSON Schema, sizes, preview image, and an example agent prompt with its render call. |
list_brand_kits | List brand kits | yes | no | — | no | Your brand kits with id, name, version, and whether a logo is set. |
get_brand_kit | Get brand kit | yes | no | — | no | A brand kit's colors, fonts, and logo URLs; omit the id for your default kit. |
preview_render | Preview a render (watermarked) | yes | no | — | no | A watermarked preview returned inline; never spends credits. |
render | Render an image | no | no | yes | no | Renders a PNG, JPEG, or WebP and returns its signed URL; repeating the same inputs returns the cached image. |
get_usage | Get usage and remaining credits | yes | no | — | no | Renders in the current period; remaining credits once billing is connected. |
There are no tools that create, change, or delete brand kits; manage brand kits with the REST API.
What render returns
- A short text summary with the URL.
structuredContentwith the same fields as the RESTPOST /v1/rendersresponse, pluswatermarkedandembedded.- A
resource_linkto the signed image URL. - The image itself, inline, when it is at most 1200×630 pixels and 1 MiB, so clients such as Claude can show it in the conversation. Larger images are returned as a URL only.
Parameter mistakes come back as tool results with isError: true and a message that names the field and the next step (for example, call get_template), not as protocol errors.
Protocol details
| Item | Value |
|---|---|
| Transport | Streamable HTTP, single POST /mcp endpoint, always application/json (no SSE stream) |
| Protocol versions | 2026-07-28 (stateless, server/discover) and the initialize-based 2025-11-25, 2025-06-18, 2025-03-26, 2024-11-05 |
| Sessions | None; no Mcp-Session-Id is issued |
| Request size | 256 KB |
| Browser requests | A cross-origin Origin header is rejected with 403 |
GET /mcp / DELETE /mcp | 405 / 204 |
| Authentication errors | 401 with WWW-Authenticate: Bearer resource_metadata="…/.well-known/oauth-protected-resource" |
| Rate limit | Shared with REST: 60 requests per 60 seconds per key |
Billing note
The render tool description says it spends one credit per new render, because that is how billing will work. In early access billing is not connected and every result reports credits_charged: 0.
Updated 2026-10-30 · Markdown version
Caching and versions
How VectorFast Render caches renders, what goes into the cache key, how template and brand kit versions work, and how idempotency keys make retries safe.
Template reference
The eight official VectorFast Render templates with their sizes, required parameters, and links to full parameter tables, examples, and agent prompts.