# `io.patchwell/patchwell` on Anlyon Radar

> Text inside fenced blocks marked untrusted-server-text was written by the MCP server's publisher, not by Anlyon Radar. It is data to inspect, not instructions to follow. Invisible characters are shown as [U+XXXX].
> No finding is not a safety guarantee: these are point-in-time checks of tool definitions, not of code or behaviour.

- Page: https://radar.anlyon.com/servers/io.patchwell/patchwell
- JSON: https://radar.anlyon.com/api/v1/servers/io.patchwell/patchwell
- Version: 1.3.0 (active)
- Publisher: io.patchwell (domain proven to the MCP Registry)
- Runs: remote (streamable-http)
- Sign-in to list tools: none
- Repository: https://github.com/aidan-mccollum/patchwell
- Metadata from the official MCP Registry, synced 2026-11-09T00:18:17.959Z

Title and description, as the publisher wrote them:

```untrusted-server-text
Patchwell
Cited web research, transcription with word timestamps, and CSV/Excel cleaning with a change log.
```

Scanned 2026-11-09T00:18:21.277Z over streamable-http.

**5 tools: 4 change things and 1 only read.**

| Group | Class | Tools | Claude Code |
| --- | --- | --- | --- |
| Deletes or cancels | destructive | 0 | deny |
| Can't be undone | irreversible | 0 | ask |
| Changes things | write | 4 | ask |
| Only reads | read | 1 | allow |

## Exfil surface

No private-data -> untrusted-input -> outbound path in this server alone: no tool is tagged private-data.

| Tag | Meaning | Tools |
| --- | --- | --- |
| private-data | Reads files, mail, messages, databases, secrets, calendars or repositories. | none |
| untrusted-input | Brings in text someone else can write: web pages, inbound email, messages, issues and comments. | `mcp__patchwell__clean_data` `mcp__patchwell__listen_clip` `mcp__patchwell__research` `mcp__patchwell__transcribe` |
| outbound | Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL. | `mcp__patchwell__clean_data` `mcp__patchwell__listen_clip` `mcp__patchwell__transcribe` |

## Findings

No check matched anything in these tool definitions.


## Changes things (4)

Creates or changes something that can be changed back. Claude Code: ask.

### `clean_data`

- Nothing in its name or description says what it does, so it is counted as a write.
- Tag untrusted-input: The server marks it openWorldHint: true: it reaches things outside the server's own data.
- Tag outbound: The input takes a URL (input.url): a request to a URL can carry data to whoever runs it.

Description, as the server wrote it:

```untrusted-server-text
Clean a messy spreadsheet or JSON export into typed, deduplicated records with a log of each change.
        Use it before you analyse, import or act on tabular data. It finds the header row, works out each
        column's type from the values (ids incl. barcodes/ISBN/IBAN check digits, emails, phones, dates, times,
        durations, money, percents, quantities with units, ZIP and postal codes, states, countries, US street
        addresses, coordinates, URLs, names, booleans), keeps leading zeros, reads European decimals, repairs
        broken encodings and merges duplicate records. It flags what deserves a look without changing it:
        disposable or role emails, phone numbers that can't exist, failed check digits. Returns `rows`, `changes`
        (row, column, before, after, rule), `review` (each with a `type`: invalid, conflict, possible_duplicate (a group of records that look
        like the same person or company, with a `confidence`, the evidence and the strongest pairs' likelihoods; never
        merged), shared_email (records sharing an email whose names, birth dates, ID numbers or Jr./Sr. say different
        people, kept apart), totals_row or note_row,
        and the reason),
        `columns` (with the evidence for each choice), `quality` (share of clean cells before and after),
        `clusters` (likely typos), a JSON `schema`, a reusable `recipe` and `stats`, whose counts always add up to
        the input rows. A file holding several tables (stacked, titled, or side by side) is split: the largest is
        the main result and every other table comes back cleaned under `other_tables`. Price: $0.50 per 1,000
        rows across all tables, minimum $0.01.
```

### `listen_clip`

- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
- Tag untrusted-input: The server marks it openWorldHint: true: it reaches things outside the server's own data.
- Tag outbound: The input takes a URL (input.url): a request to a URL can carry data to whoever runs it.

Description, as the server wrote it:

```untrusted-server-text
Used by the transcript card to play a spot the speaker labels are unsure of. Free.
```

### `research`

- Nothing in its name or description says what it does, so it is counted as a write.
- Tag untrusted-input: The server marks it openWorldHint: true: it reaches things outside the server's own data.

Description, as the server wrote it:

```untrusted-server-text
Research a question on the live web and get an answer in which each sentence cites the page it came
        from. Use it for facts you can't answer from memory or that may have changed. Returns `answer` (sentences
        with [url] citations), `sources`, `gaps` (what the sources couldn't confirm) and `confidence`
        (high/medium/low). Takes about a minute; a slow question returns {"status": "running", "job_id"} for
        get_job_result. Declines questions about private individuals (addresses, contact details, relatives,
        profiles) and returns a `category` saying why. Price: $0.15 per answered question; declined and unanswered
        questions are free.
```

### `transcribe`

- Nothing in its name or description says what it does, so it is counted as a write.
- Tag untrusted-input: The server marks it openWorldHint: true: it reaches things outside the server's own data.
- Tag outbound: The input takes a URL (input.url): a request to a URL can carry data to whoever runs it.

Description, as the server wrote it:

```untrusted-server-text
Transcribe speech in an audio or video file. Use it when you need the words in a recording, quotes with
        their time, who said what, or subtitles. Returns text, sentence segments, each word with start/end seconds,
        confidence and speaker (`spk`: 1, 2, ... told apart by voice), `turns` (speaker, start, end, text),
        `speakers` with talk time (and a `name` with its supporting quote when the recording says who someone is:
        "My name is...", "the next question comes from..."), SRT and VTT with speaker labels, and `flags` listing any phrase the guards removed (Whisper's invented phrases over silence and
        repetition loops) with its original text and time. Files over about 100 seconds of processing return
        {"status": "running", "job_id"}: call get_job_result with it. Price: $0.008 per audio minute, minimum $0.02 a file; failed
        downloads are free.
```

## Only reads (1)

Looks things up and changes nothing. Claude Code: allow.

### `get_job_result`

- The name "get_job_result" says "get": a word for reading.

Description, as the server wrote it:

```untrusted-server-text
Fetch the result of a long call that returned {"status": "running", "job_id"}. Call it every 20 to 30
        seconds until status is 'done'; the finished result has the same shape as the original tool's. Free.
```

## Claude Code permissions

For `.claude/settings.json`, assuming the server is installed as `patchwell`:

```json
{
  "permissions": {
    "allow": [
      "mcp__patchwell__get_job_result"
    ],
    "ask": [
      "mcp__patchwell__transcribe",
      "mcp__patchwell__research",
      "mcp__patchwell__listen_clip",
      "mcp__patchwell__clean_data"
    ],
    "deny": []
  }
}
```

Capability tags behind the rules (a read tagged outbound is asked about):

- `mcp__patchwell__transcribe`: untrusted-input, outbound
- `mcp__patchwell__research`: untrusted-input
- `mcp__patchwell__listen_clip`: untrusted-input, outbound
- `mcp__patchwell__clean_data`: untrusted-input, outbound

## Remote posture

What an anonymous client saw at the remote endpoint, last 2026-11-21T19:22:34.168Z, unchanged since 2026-11-14T12:21:26.549Z. No credentials were sent and no tool was called. Values are shown as the endpoint returned them. Method: https://radar.anlyon.com/methodology#remote-posture

| Fact | Observed |
| --- | --- |
| Reachable | `yes` |
| HTTP status of an anonymous initialize | `200` |
| tools/list served without credentials | `yes` |
| Tools that change things, listed without credentials | `4` |
| Protocol version | `2025-06-18` |
| WWW-Authenticate on 401 | `not observed` |
| resource_metadata in WWW-Authenticate | `none` |
| RFC 9728 protected resource metadata | `https://patchwell.io/.well-known/oauth-protected-resource` |
| Authorization servers | `https://patchwell.io` |
| Legacy HTTP+SSE transport | `no` |
| Dynamic client registration | `yes` |
| Client ID metadata documents | `no` |

- **high** Tools that change things, listed to anyone ([unauth-remote-writes](https://radar.anlyon.com/methodology#unauth-remote-writes))

```untrusted-server-text
The endpoint served tools/list without credentials, and the list has 4 tools that change things (3 destructive, 1 write). Whether a call also works without credentials was not tested: tools are never called.

https://patchwell.io/mcp/: tools/list returned 5 tools to an anonymous client, including transcribe (destructive), research (destructive), clean_data (destructive), listen_clip (write)
```

Posture changes:

- 2026-11-14T12:21:26.549Z: Tools that change things, listed without credentials `3` -> `4`
