API reference

Quickstart#

Create an account to get an API key, then mark your first file. The marked file is written to render.marked.png, and the proof details print as response headers.

curl https://markproof.co/v1/mark \  -H "Authorization: Bearer $MARKPROOF_KEY" \  -F file=@render.png \  -F generator_name="your-app" \  -F model_name="your-model-v2" \  -o render.marked.png -D -

Authentication#

Send your key as a bearer token on every request: Authorization: Bearer lt_live_…. Keys are shown once, when they're created. If you lose one, generate a new key from your dashboard. The old key stops working immediately.

Keep keys on your server. Never ship one in browser or mobile app code.

SDKs & integrations#

Official packages wrap the API below, so you don't have to build requests by hand.

Node.js and Python

npm install markproof import { readFileSync } from "node:fs";import { Markproof } from "markproof"; const markproof = new Markproof({ apiKey: process.env.MARKPROOF_API_KEY });const result = await markproof.mark(  { data: readFileSync("render.png"), filename: "render.png" },  { generatorName: "your-app" },);// result.file is the marked PNG; result.watermarkCode, result.verifyUrl …

pip install markproof import osfrom markproof import Markproof markproof = Markproof(os.environ["MARKPROOF_API_KEY"])result = markproof.mark("render.png", generator_name="your-app")# result.file is the marked PNG; result.watermark_code, result.verify_url …

Vercel AI SDK

Wrap your image model once and every generateImage result is marked before it reaches your code. The model's provider and ID are recorded in the C2PA manifest automatically. If marking fails, the call fails too, so an unmarked image is never returned by mistake. For audio from generateSpeech, use markGeneratedFile(audio).

npm install markproof-ai import { generateImage, wrapImageModel } from "ai";import { openai } from "@ai-sdk/openai";import { markproofImageMiddleware } from "markproof-ai"; const model = wrapImageModel({  model: openai.image("gpt-image-1"),  middleware: markproofImageMiddleware({ generatorName: "your-app" }),}); // Every image comes back already marked.const { images, providerMetadata } = await generateImage({ model, prompt: "…" });// providerMetadata.markproof.images[i] → watermarkCode, verifyUrl, …

AI agents (MCP)

markproof-mcp is a Model Context Protocol server. It gives agents in Claude, Cursor and other MCP clients three tools: mark_file (writes name.marked.png beside the original), verify_file and get_asset. Add it to your client's MCP config, or with Claude Code run claude mcp add markproof --env MARKPROOF_API_KEY=lt_live_… -- npx -y markproof-mcp.

{  "mcpServers": {    "markproof": {      "command": "npx",      "args": ["-y", "markproof-mcp"],      "env": { "MARKPROOF_API_KEY": "lt_live_…" }    }  }}

LangChain and LlamaIndex

Both frameworks load MCP servers as agent tools, so markproof-mcp works with them directly. Pass your key through env: MCP clients often don't forward your whole environment to the server process.

pip install "langchain[mcp]" import osfrom langchain.agents import create_agentfrom langchain.mcp import MCPAdapter config = {"mcpServers": {"markproof": {    "command": "npx",    "args": ["-y", "markproof-mcp"],    "env": {"MARKPROOF_API_KEY": os.environ["MARKPROOF_API_KEY"]},}}} async with MCPAdapter(config) as adapter:    tools = await adapter.list_tools()  # mark_file, verify_file, get_asset    agent = create_agent("anthropic:claude-sonnet-5-5", tools)    await agent.ainvoke({"messages": [{"role": "user", "content": "Mark render.png with Markproof."}]})

pip install llama-index-tools-mcp import osfrom llama_index.core.agent.workflow import FunctionAgentfrom llama_index.tools.mcp import BasicMCPClient, McpToolSpec client = BasicMCPClient(    "npx", args=["-y", "markproof-mcp"],    env={"MARKPROOF_API_KEY": os.environ["MARKPROOF_API_KEY"]},    timeout=120,  # the default 30s can be too short for a first, cold-start call)tools = await McpToolSpec(client=client).to_tool_list_async()agent = FunctionAgent(tools=tools, llm=llm)  # any LlamaIndex LLM

ComfyUI

The Markproof node pack adds Mark & Save, which you use in place of Save Image (re-encoding would strip the C2PA manifest), and Verify Image. Clone it into ComfyUI/custom_nodes; a ComfyUI-Manager listing is in review. Set MARKPROOF_API_KEY in the environment before starting ComfyUI.

Mark a file#

POST/v1/mark

Adds an invisible watermark and a signed C2PA manifest to an image or audio file. Send the file as multipart/form-data, or send JSON with a url for Markproof to fetch it from.

Response

With response_format=file, the body is the marked file and the details come back as headers: X-Asset-Id, X-Watermark-Code, X-Verify-Url, X-C2pa-Manifest-Label and, for images, X-Watermark-Robust. Images are returned as PNG and audio as WAV, so the watermark isn't degraded by re-compression.

With response_format=url, you get JSON. The download link is temporary, and the stored file is deleted after 30 days.

Watermark robustness

Every image is checked after marking: the watermark must still decode after JPEG re-compression, a 50% resize, a 10% crop and WebP conversion. Markproof uses the faintest watermark that passes. For rare, very smooth or featureless images no setting passes, and watermark_robust comes back false. The file is still marked and signed, but its watermark may not survive editing, so rely on the C2PA manifest for those files.

curl https://markproof.co/v1/mark \  -H "Authorization: Bearer $MARKPROOF_KEY" \  -H "Content-Type: application/json" \  -d '{"url": "https://cdn.example.com/render.png", "response_format": "url"}'

{  "asset_id": "5e847be7-7d96-4808-85c3-7cbeb5233351",  "watermark_code": "ARN26QVH",  "kind": "image",  "verify_url": "https://markproof.co/v/ARN26QVH",  "c2pa_manifest_label": "urn:c2pa:c600893b-eab4-4ae3-919f-e5da47cc7f50",  "watermark_robust": true,  "file_url": "https://…signed-download-url…"}

Verify a file#

POST/v1/verify

Checks any image or audio file for a Markproof watermark and a C2PA manifest. Accepts file or url, like /v1/mark. A file with neither is still a 200, because “not marked” is a valid answer, not an error.

curl https://markproof.co/v1/verify \  -H "Authorization: Bearer $MARKPROOF_KEY" \  -F file=@render.marked.png

{  "watermark": { "detected": true, "watermark_code": "ARN26QVH", "confidence": null },  "c2pa": {    "present": true,    "signature_valid": true,    "trusted": false,    "manifest_label": "urn:c2pa:c600893b-eab4-4ae3-919f-e5da47cc7f50"  },  "matched_asset": {    "id": "5e847be7-7d96-4808-85c3-7cbeb5233351",    "watermark_code": "ARN26QVH",    "kind": "image",    "generator_name": null,    "model_name": null,    "c2pa_manifest_label": "urn:c2pa:c600893b-eab4-4ae3-919f-e5da47cc7f50",    "created_at": "2026-10-27T00:52:35.390961+00:00"  },  "summary": "Watermark detected code ARN26QVH, matches our record. C2PA manifest present and valid, signed by an untrusted certificate."}

Look up an asset#

GET/v1/assets/{id}

Returns the record for a file you marked, using the asset_id from /v1/mark. Any key on the same project works. Assets from other projects return 404.

curl https://markproof.co/v1/assets/5e847be7-7d96-4808-85c3-7cbeb5233351 \  -H "Authorization: Bearer $MARKPROOF_KEY"

Public verify page#

Every marked file gets a public page at https://markproof.co/v/{watermark_code}, returned as verify_url. Anyone can open it without an account. It shows the media type, generator, marking date and C2PA manifest label, and nothing else about your account. Anyone can also check a file directly with the free checker.

Limits#

Errors#

Errors return JSON with an error message, for example {"error": "Monthly quota exceeded (250/250)."}.