Patchwell

Patchwell

Cited web research, transcription with word timestamps, and spreadsheet cleaning with a change log, for AI agents.
Remote MCP server and plain HTTP API. Pay per call, no subscription.

patchwell.io · Quickstart · API reference · MCP Registry

---

Tools

Tool	What it returns	Price
research	An answer where each sentence cites the page it came from, what the sources couldn't confirm, and a confidence grade. Topics, markets, products, companies, regulations, published research. Won't look up private people.	$0.15 per answered question. Declined or unanswered: free
transcribe	Text, word-level start/end times and confidence, who said each word (Speaker 1, 2, ..., or their name when the recording says it, with the quote as evidence), speaker turns, sentence segments, SRT and VTT with speaker tags. Audio or video URLs up to 3 hours. When Whisper invents a phrase or loops on a line, you get a flag with the original text and timestamps.	$0.008 per audio minute
clean_data	Clean rows from a CSV, Excel or JSON file, plus a log of every changed value, a review list, detected column types, a JSON Schema and a reusable recipe.	$2 per 1,000 rows
get_job_result	The result of a long job (anything over about 100 seconds returns a job_id).	Free

Measured on real files: a 45-minute earnings call with 7 speakers took about 90 seconds, speaker labels included. On 24 public recordings (AMI meetings, VoxConverse debates and news, Earnings-21 calls), 97% of words were credited to the right speaker. A messy 352-row customer file took 2.2 seconds to clean, with 45 duplicates merged and 1,435 values fixed, each one logged. A research question takes about a minute.

Get a key

Free $1 key, no card, one per email:

curl -X POST https://patchwell.io/v1/credits/trial \
  -H "Content-Type: application/json" -d '{"email": "[email protected]"}'

Or buy credit at patchwell.io. Keys look like pw_live_....

Connect

Claude (web and desktop) or ChatGPT

Add a custom connector with the address https://patchwell.io/connect/. The app opens a Patchwell page where you paste your key or get a free one, then sends you back. In Claude it's under Customize, then Connectors, then Add. In ChatGPT, use developer mode.

Claude Code

claude mcp add --transport http patchwell https://patchwell.io/mcp/ --header "Authorization: Bearer YOUR_KEY"

Cursor, Windsurf and other clients that take JSON config

{
  "mcpServers": {
    "patchwell": {
      "type": "http",
      "url": "https://patchwell.io/mcp/",
      "headers": { "Authorization": "Bearer YOUR_KEY" }
    }
  }
}

Claude Desktop (through the mcp-remote bridge)

{
  "mcpServers": {
    "patchwell": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://patchwell.io/mcp/", "--header", "Authorization: Bearer YOUR_KEY"]
    }
  }
}

Plain HTTP

curl https://patchwell.io/v1/research \
  -H "Authorization: Bearer $PATCHWELL_KEY" -H "Content-Type: application/json" \
  -d '{"ask": "What does Stripe charge for online card payments in the US?"}'

What a research answer looks like

Real output, shortened:

{
  "answer": "Stripe charges 2.9% + 30 cents for standard online domestic card payments in the US [https://www.swipesum.com/insights/guide-to-stripe-fees-rates-for-2025][https://checkoutpage.com/blog/stripe-processing-fees]. PayPal's standard merchant fee is 3.49% + 49¢ per online transaction [https://merchantinsiders.com/blogs/paypal-fees/]. ...",
  "confidence": "medium",
  "gaps": ["2 claims left out because the sources found did not clearly support them."],
  "billing": { "price": 0.15 }
}

Before you see the answer, a separate step compares each claim with the text of the page it cites. It drops the claims that page doesn't support and counts them in gaps.

Also on Apify

The same tools run as Apify Actors, billed through your Apify account: research · transcribe · clean data

Built for automations

• Retries never charge twice: send an Idempotency-Key header; a retry gets the first response back.

• Webhooks: add callback_url and get the result POSTed when it's done, signed with HMAC-SHA256 and retried for about two hours.

• Presets: save a configuration once (PUT /v1/presets/<name>) and send "preset": "<name>".

• Exports: POST /v1/export?format=docx (or srt, vtt, txt, md, csv) turns a transcript into a file.

• Speaker labels you can check: every word carries spk_p_wrong, the measured chance its speaker is wrong, and speaker_review lists the spots worth a listen with links that open the file at that second.

• Stable results: every result has schema_version; within v1, fields are only added.

• Talk-time analytics: each speaker's share, pace, turns, longest stretch, questions and interruptions.

• Rename speakers in the Claude card, or with POST /v1/rename and speaker_names on exports.

• Templates: n8n workflows (tested inside n8n) and a schedulable podcast-to-notes script at patchwell.io/automations.

How these numbers were measured, and where the tools fall short: patchwell.io/trust.

Privacy and limits

• Patchwell processes audio and files in temporary containers and discards them. It keeps no copies.
• Patchwell keeps research questions for 30 days, with contact details removed, to improve accuracy. Switch that off per key at patchwell.io/privacy, or per request with the header X-Patchwell-Retain: off.
• Research declines questions about private individuals (addresses, phone numbers, relatives, profiles). If a key keeps asking, Patchwell pauses it.
• The tools refuse URLs that point at private or internal network addresses.

About this repository

This repo holds the MCP Registry entry (server.json) and these docs. The service runs at patchwell.io.

Questions: [email protected]