Developer interfaces

MCP server (PDF processing)

A standard MCP (Model Context Protocol) Streamable HTTP endpoint with 9 PDF tools (merge / split / rotate / remove pages / compress / watermark / page numbers / metadata / form flatten). Anonymous, no API key; ≤10 MB and ≤200 pages per file; only successful calls consume the daily free quota, failures are refunded automatically (up to a daily refund allowance), quotas reset at 00:00 UTC.

Endpoint:POST https://pdfsuite.zalize.com/mcp

Client config (Claude / Codex / Cursor):

{
  "mcpServers": {
    "pdf-suite": {
      "type": "http",
      "url": "https://pdfsuite.zalize.com/mcp"
    }
  }
}

Full tool list and input conventions on the MCP server page →

llms.txt (LLM-friendly site map)

A site summary and page index following the llmstxt.org convention, for LLMs and agents to understand the site:

https://pdfsuite.zalize.com/llms.txt

Public stats endpoint

GET /api/public-stats returns an anonymous aggregate of successful tool runs over the last 30 days (the same source as the homepage trust number; no user-level data). The homepage hides the number below 100 samples.

GET https://pdfsuite.zalize.com/api/public-stats
→ { "window_days": 30, "success": <number> }

What does not exist (honest boundaries)

• No general REST API: programmatic PDF processing is offered over the MCP endpoint only.
• No API keys or developer accounts: MCP calls are anonymous and share the same per-IP daily usage pool as the website, with a stricter cap on the MCP side — once the IP’s combined usage reaches the per-tool daily limit, MCP calls are rejected (the website counts per device, with an IP pool cap of 4× the daily limit).
• Rendering-based conversions (PDF to images, OCR) and AI translate / summarize / chat have no programmatic interface — they run locally in the browser or on the signed-in website.
• The browser extension is a shortcut to the web tools and has no programmable surface.