fynepdf logo
Developer platform. Opening 2027.

The PDF API and MCP server for apps and agents.

Everything fynepdf.com does, as a REST API, a remote MCP server, and plugins for Claude, ChatGPT, Muse and Grok. Deterministic results, with a receipt for every step.

Free to join

Early access goes to the list first, in order. One email when it opens, nothing else.

Servers in the EU
Files deleted after processing
Never used to train a model

One catalog. Call it your way.

A job is a file and up to eight steps. Send it from code, from an agent over MCP, or from a plugin in the assistant you already use.

Preview, not final
// One job, up to eight steps, one result
POST https://api.fynepdf.com/v1/jobs
Authorization: Bearer fp_live_........

{
  "input": "asset_8f3k2q",
  "steps": [
    { "action": "compress", "targetSize": "2MB" },
    { "action": "pageNumbers", "position": "bottom-right" },
    { "action": "convert", "to": "pdfa" }
  ],
  "webhook": "https://example.com/hooks/fynepdf"
}

// 202 Accepted
{ "job": "job_01hz", "status": "queued", "quote": { "credits": 3 } }

Four doors onto the same room

The product, the API, the MCP server and the plugins run one set of operations on one set of servers. A fix or a new operation lands in all four at once.

Live

fynepdf.com

Tell Fyne what to do. It plans the steps, you press Run.

2027

REST API

Upload, quote and run from code. Poll, or take a webhook.

2027

MCP server

Remote, with OAuth. For any agent you build or run.

2027

Plugins

Claude, ChatGPT, Muse and Grok. Turn it on, then ask.

Three calls, start to finish

1

Upload

Ask for a signed upload address, put the file there, finalize. You get an asset id. Large files go straight to storage, never through the API.

POST /v1/assets
2

Quote, then run

Send the steps. A quote comes back first if you want one. Each step feeds the next; heavy work moves to a queue on its own.

POST /v1/jobs
3

Collect the receipt

Poll or take the signed webhook. The result carries the file, the pages and sizes before and after each step, and which step failed if one did.

GET /v1/jobs/{id}

What you can call

Deterministic by default. The few operations that use a model are marked, metered separately, and never run unless you ask for them.

Compress to a size

Say under 2 MB. The target is checked on the final file.

Merge and split

Combine files, pick pages across them, or split by range.

Pages

Delete, keep, reorder, rotate, number, watermark, grayscale.

OCR

Scans become searchable PDFs. 13 languages ready, more on request.

PDF to Word

Editable documents that keep their layout and fonts.

PDF to Excel

Tables out as spreadsheets, with numbers that still add up.

Office and web to PDF

Word, Excel, PowerPoint, Markdown, HTML and images in.

PDF to images and Markdown

Pages as PNG or JPEG, text as Markdown for your pipeline.

PDF/A, lock, flatten, repair

Archive files, lock or unlock, flatten forms, fix damage.

Sign

Stamp a signature image on a named spot or exact coordinates.

Redact

AI to find, metered

Remove text and images for real. Mark them, or let AI find them.

Ask with citations

AI, metered

Answers with page numbers, tables as CSV, totals checked.

Every tool on fynepdf.com is on the list for launch. Heavy runs move to a job queue on their own.

A receipt for every run

A result you cannot check is a guess. Every job comes back with pages and sizes before and after each step, which step failed if one did, and a size target verified on the final file, not on a hope.

  • Up to eight steps in one job. Each step feeds the next.
  • A quote before the work, so your app knows the cost.
  • Idempotency keys and signed webhooks. A retry never runs twice.

job_01hz, done

Example
  1. 1compress8.4 MB to 1.9 MB
  2. 2pageNumbers24 pages
  3. 3convertPDF/A-2b
Size target under 2 MB: met3 credits, 3.1 s

Agents need hands, not more words

A model can describe what should happen to a PDF. It cannot compress one, OCR one, or prove a redaction removed the text. The MCP server and the plugins give any agent those hands, with the same receipts your app gets.

  • Deterministic by default

    A FynePDF model runs only when a step is an AI step.

  • Linked to your account

    OAuth to a FynePDF account. No purchase tools for the agent.

  • Six small tools

    List actions, upload, quote, run, status, result. Plugins use the same six.

Free to build. Pay for what runs.

Developer billing is separate from the fynepdf.com plans. Credit rates are published at launch; the shape is settled now.

Free allowance

A monthly allowance for building and testing. No card to start.

Prepaid credits

Credit packs, bought once and spent on jobs. No auto top-up.

Teams and volume

Volume rates and a data processing agreement, set up with you.

Frequently Asked Questions

When does the FynePDF API open?

In 2027, once the product side has run in public long enough to trust. The REST API, the MCP server and the plugins open on the same day. The waitlist gets access first, in the order people joined.

What can the FynePDF API do with a PDF?

Everything fynepdf.com does, in jobs of up to eight steps: compress to a target size, merge and split, delete, reorder, rotate and number pages, watermark, OCR scans, convert PDFs to Word, Excel, images or Markdown, turn Word, Excel, PowerPoint, HTML and images into PDFs, make PDF/A, lock, unlock, flatten, repair, sign and redact. Asking questions with page citations is an AI step, metered separately.

How much will the FynePDF API cost?

Every account gets a free monthly allowance for building and testing, with no card needed. After that you buy prepaid credit packs and spend them on jobs, with no subscription and no auto top-up. Credit rates are published at launch.

Which assistants get a FynePDF plugin?

Claude, ChatGPT, Muse and Grok at launch. Any other MCP client, including agents you build, connects through the remote MCP server, which is what the plugins use underneath.

Does an API request ever call an AI model?

Only when you ask for an AI step, such as finding what to redact or extracting a table with AI. Those are marked in the catalog and metered separately. Everything else is deterministic and gives the same output for the same input.

What happens to files sent through the API?

They are processed on FynePDF's servers in Frankfurt, in the EU. AI steps also send the document to Google's Gemini API to produce the answer. Files are deleted automatically after processing and never used to train a model. The exact retention window is published with the API terms.

Is the API the same engine as the fynepdf.com website?

Yes. The product, the API, the MCP server and the plugins call one action catalog on one set of servers. A fix or a new operation lands everywhere at once.
Developer platform

Be first to build on FynePDF.

Early access, the docs as they are written, and a say in which operations ship first. One email when it opens.

Join the waitlistExplore the tools
Early accessWaitlist open

2027launch

API, MCP server and plugins, on the same day

  • Access before the public launch
  • Free monthly allowance from day one
  • Docs and changelog as they are written
  • A direct line to the founder
Join the waitlist