# harald (pypi · harald)

helps coding agents work out the bluetooth le and classic protocol of gadgets you own

- Trust score: 65/100 (medium)
- Registry status: active
- Liveness: live
- Owner verified: no
- Last scored: 2026-11-04

## Components

- pypi · `harald`: 65/100 (this document), [markdown](https://verifymcp.io/servers/k0nnect-harald/harald.md), [page](https://verifymcp.io/servers/k0nnect-harald/harald)

## Channel facts

- Registry: `pypi`
- Package: `harald`
- Version: `0.2.2`
- Transport: `stdio`

## Trust breakdown

How this component scores in each security and reliability category. Every signal is checked automatically from public evidence about the published package, including repeated runs of it in an isolated sandbox, and we only credit what we can confirm. Scores are 0–100 per category. Scoring method: https://verifymcp.io/docs/scoring (what has changed: https://verifymcp.io/docs/scoring/changelog)

Scored 2026-11-04.

- **Supply Chain Security**: 100/100
  - No malware found by supply-chain analysis.
  - No known CVEs affecting this package version or its production dependencies.
  - Runs hatchling.build at install time, a recognised build step with no custom scripting around it.
  - 1 of 48 dependencies flagged as unhealthy.
- **Provenance & Transparency**: 45/100
  - Source repository is publicly reachable at the declared URL.
  - Provenance check failed: no build-provenance attestation is published.
  - Clear OSI-approved license (MIT).
  - Actively maintained (last published 1 days ago).
  - Disclosure check failed: no security disclosure policy was found in the source repository.
- **Schema Quality & AI Usability**: 65/100
  - AI-judged instruction clarity (good).
  - Tool/resource definitions use about 1801 tokens (~100/item across 18 items; 18 tools + 0 resources), lean.
  - Usage-examples check failed: none of the tools include examples.
- **Stability & Change Management**: 0/100
  - Stability not yet verified: not enough scan history yet (needs a 30-day window).
- **Tool Coverage**: 71/100
  - 100% of tools have a non-trivial description (not blank, and not just the tool's name).
  - 0% of tool parameters carry a description.
  - Structured output schemas are declared (100% of tools); any adoption earns full credit.
- **Tool Safety**: 75/100
  - No prompt-injection markers were found in the server instructions, tool names or descriptions we captured.
  - 0 of 1 tool(s) whose name or description implies an irreversible operation declare an MCP destructiveHint annotation; "classic_send" implies "send" and declares no destructiveHint at all, which the MCP spec reads as destructive by default.
  - An AI judge read all 19 captured unit(s) of tool text and found none that tries to manipulate the model reading it.
- **Capabilities**: 100/100
  - Implements a current MCP spec version (2026-07-28).

**Unverified: 1 category.** A category scored 0 because we could not verify it: a data source with nothing on this package, evidence we could not reach, or a check we could not run. We only credit what we can confirm.

## Install

### How do I install the harald MCP server?

harald runs locally as a PyPI package, launched with uvx harald. Ready-made configuration for Claude, Cursor, VS Code, Codex and 5 more is on this page, copied from each client's own documentation.

### Claude

```bash
claude mcp add k0nnect-harald -- uvx harald
```

### Cursor

```json
{
  "mcpServers": {
    "k0nnect-harald": {
      "command": "uvx",
      "args": [
        "harald"
      ]
    }
  }
}
```

### VS Code

```json
{
  "servers": {
    "k0nnect-harald": {
      "command": "uvx",
      "args": [
        "harald"
      ]
    }
  }
}
```

### Codex

```bash
codex mcp add k0nnect-harald -- uvx harald
```

### opencode

```json
{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "k0nnect-harald": {
      "type": "local",
      "command": [
        "uvx",
        "harald"
      ],
      "enabled": true
    }
  }
}
```

### OpenClaw

```bash
openclaw mcp add k0nnect-harald --command uvx --arg harald
```

### Hermes

```yaml
mcp_servers:
  k0nnect-harald:
    command: "uvx"
    args: ["harald"]
```

### Netclaw

```json
{
  "McpServers": {
    "k0nnect-harald": {
      "Transport": "stdio",
      "Command": "uvx",
      "Arguments": [
        "harald"
      ]
    }
  }
}
```

### Vellum

```bash
assistant mcp add k0nnect-harald -t stdio -c uvx -a harald
```

### Other

```json
{
  "mcpServers": {
    "k0nnect-harald": {
      "command": "uvx",
      "args": [
        "harald"
      ]
    }
  }
}
```

## Changelog

Every change recorded for this component, newest first. Days that predate change tracking, or that we cannot explain, say so: "we were watching and nothing happened" and "we were not watching" are different claims.

### 2026-11-03 (score 65, +15)

- [security improvement] Malware scan: unverified → pass
- [security improvement] Known CVEs: partial → pass
- [functional] Dependency health: partial → 1.00

### 2026-11-02 (score 50)

First indexed and scored.

## MCP tools (18)

### `scan` (~48 tokens)

scan for bluetooth le devices for `timeout` seconds, strongest signal first.

manufacturer_data maps each company id (as 0x-hex) to its payload hex.

Input parameters:

- `timeout` (number)

Output parameters:

- `result`

### `connect` (~44 tokens)

connect to a device and list its gatt services and their characteristics.

use the characteristic uuids listed here with read_char, write_char and subscribe.

Input parameters:

- `address` (string, required)

Output parameters:

- `result`

### `read_char` (~34 tokens)

read a characteristic once; returns its value as hex.

Input parameters:

- `address` (string, required)
- `char_uuid` (string, required)

Output parameters:

- `result`

### `write_char` (~120 tokens)

write hex bytes to a characteristic; returns the writes left.

refused unless the human armed this device by running `harald arm <address>`
in their own terminal. each write spends one from that budget, and
firmware-update services stay blocked. by default it writes without response
when the characteristic allows that, and with response otherwise.
response=true waits for the device to acknowledge the write; false does not.

Input parameters:

- `address` (string, required)
- `char_uuid` (string, required)
- `data_hex` (string, required)
- `response`

Output parameters:

- `result`

### `subscribe` (~54 tokens)

collect notifications from a characteristic for `seconds`.

returns one entry per notification: seconds since the start and the value as hex.

Input parameters:

- `address` (string, required)
- `char_uuid` (string, required)
- `seconds` (number, required)

Output parameters:

- `result`

### `open_capture` (~292 tokens)

show the att and rfcomm timeline of a phone capture (.pklg or btsnoop log).

one line per event with time, acl connection, direction, op and value hex. att
events show their handle and uuid; rfcomm events (rfcomm data, rfcomm_open and
rfcomm_close) show their server channel, e.g. ch=9. any warnings come first and
a `showing a–b of n events` footer comes last. a capture can hold several
connections, each with its own handles. page with limit and offset. handle (an
integer, e.g. 18 for h=0x0012) keeps only att events on that handle, channel
(e.g. 9) only rfcomm events on that channel, conn (e.g. 64 for c=0x0040) only
events on that connection, and offset then counts those filtered events. for a
bluetooth classic gadget, capture its android app in a btsnoop log: an iphone
app reaches classic gadgets only through apple iap2, which harald shows as
rfcomm data but can't unwrap yet. relative paths resolve against the server's
working directory.

Input parameters:

- `channel`
- `conn`
- `handle`
- `limit` (integer)
- `offset` (integer)
- `path` (string, required)

Output parameters:

- `result`

### `diff` (~69 tokens)

compare hex payloads byte by byte.

returns how many payloads have each length, the byte at every position that is
the same in all of them (over the shortest), the positions that vary, and
whether the set looks like block-cipher output.

Input parameters:

- `payloads_hex` (array, required)

Output parameters:

- `result`

### `guess_checks` (~83 tokens)

name the checksums and length fields one hex payload is consistent with.

checks sum8 and xor8 in the last byte, and the byte before a trailing end
marker, over the bytes after a leading start marker; crc16-ccitt in the last
two; and a length byte at position 0 or 1.

Input parameters:

- `payload_hex` (string, required)

Output parameters:

- `result`

### `try_keys` (~120 tokens)

decrypt hex payloads with each candidate aes key, in ecb and in cbc with a
zero iv, best-scoring first. samples holds the first plaintexts as hex.

the score only tells keys apart across several distinct payloads: with one
payload, or identical copies, every key scores 1.0. keys that are not 16, 24
or 32 bytes and payloads that are not a multiple of 16 bytes are skipped.

Input parameters:

- `keys_hex` (array, required)
- `payloads_hex` (array, required)

Output parameters:

- `result`

### `analyze_apk` (~172 tokens)

statically analyze a vendor android app (.apk, .xapk or .apks) already on disk.

detects the framework. for a native app it runs jadx and lists the uuids
(gatt and rfcomm) in gatt_uuids and file:line candidates (uuids, aes keys,
base64 blobs, command builders, rfcomm socket calls) to open under jadx_dir,
the app's own package first and third-party libraries skipped.
each list keeps at most 50 per kind; totals has the full counts, and when
anything was cut a warning names the file under jadx_dir that lists them all.
flutter and react-native apps keep their ble code compiled, so the warnings
point to capture analysis instead. harald never downloads apps.

Input parameters:

- `path` (string, required)

Output parameters:

- `result`

### `list_devices` (~23 tokens)

list the gadgets in harald's device library, with status and summary.

Output parameters:

- `result`

### `load_device` (~31 tokens)

the device library entry for one slug, including any app-wide key and cipher.

Input parameters:

- `slug` (string, required)

Output parameters:

- `result`

### `read_notes` (~52 tokens)

the working notes for a device, or an empty string when there are none.

notes live in harald-work/<slug>/notes.md, relative to the server's working
directory.

Input parameters:

- `slug` (string, required)

Output parameters:

- `result`

### `save_note` (~68 tokens)

append a timestamped note to harald-work/<slug>/notes.md.

the path is relative to the server's working directory. slugs use lowercase
letters, digits and hyphens, e.g. demo-lamp.

Input parameters:

- `slug` (string, required)
- `text` (string, required)

Output parameters:

- `result`

### `classic_devices` (~46 tokens)

list the bluetooth classic devices paired with this computer.

connected says whether each is linked to this computer now. use an address
from here with classic_services, classic_send and classic_listen.

Output parameters:

- `result`

### `classic_services` (~103 tokens)

list a paired classic device's service records from this computer's cache.

each record has a name, service-class uuids and an rfcomm channel (null when
rfcomm cannot reach it). protected channels are standard profiles such as
hands-free and headset, which harald never opens; firmware_update channels
stay blocked for writes. refresh=true asks the device again instead of
trusting the cache.

Input parameters:

- `address` (string, required)
- `refresh` (boolean)

Output parameters:

- `result`

### `classic_send` (~226 tokens)

send hex messages to a classic device's rfcomm channel and collect its answers.

refused unless the human armed this device by running `harald arm <address>`
in their own terminal. each message spends one armed write, a query included,
because classic has no read. all messages go over one connection, in order,
with `listen` seconds (at most 30) of listening after each, so handshakes and
acks can be sent as messages. service is an rfcomm service uuid or channel number from
classic_services; standard profiles and channels missing from
classic_services are refused, and firmware-update channels stay blocked.
received holds each chunk with the message number it followed (0 = before the
first), seconds since the channel opened and the value as hex. stopped
explains a budget that ran out partway, or a link that failed after the first
message; it is empty when every message was sent.

Input parameters:

- `address` (string, required)
- `listen` (number)
- `messages_hex` (array, required)
- `service` (string, required)

Output parameters:

- `result`

### `classic_listen` (~101 tokens)

open a classic device's rfcomm channel, send nothing, and collect `seconds`
(at most 30) of what it says.

needs no arm, but standard profiles are still refused. service is an rfcomm
service uuid or channel number from classic_services. returns one entry per
chunk: seconds since the channel opened and the value as hex.

Input parameters:

- `address` (string, required)
- `seconds` (number, required)
- `service` (string, required)

Output parameters:

- `result`

## Diagnostics

Captured diagnostic sections: Provenance, Install scripts, Dependencies. The full working is on the page: https://verifymcp.io/servers/k0nnect-harald/harald#diagnostics

## Score history

- 2026-11-04: 65
- 2026-11-03: 65
- 2026-11-02: 50

## Common questions

### What is the harald MCP server?

harald is an MCP server listed in the public MCP registry as io.github.k0nnect/harald. helps coding agents work out the bluetooth le and classic protocol of gadgets you own. This page covers its PyPI package (harald).

### Is the harald MCP server safe to use?

harald scores 65 out of 100 on VerifyMCP. We found no known CVEs affecting it as of 4 November 2026. That is a record of what we were able to check automatically, not an endorsement. The category breakdown on this page shows every signal behind the number, including the ones we could not confirm.

### What tools does the harald MCP server expose?

harald exposes 18 tools: scan, connect, read_char, write_char, subscribe, and 13 more. Their descriptions and schemas cost roughly 1,686 tokens of context every time the server is loaded.

### Is the harald MCP server still maintained?

harald is still listed as active in the MCP registry. We last reached this channel on 4 November 2026. Those dates come from our own scans of the registry and the channel itself, not from anything the publisher announced.

### What licence is the harald MCP server under?

harald declares the MIT licence, which is OSI-approved. That covers the source only, and says nothing about the cost of any service it calls.

## Links

- PyPI project: https://pypi.org/project/harald/
- Socket report: https://socket.dev/pypi/package/harald
- Repository: https://github.com/k0nnect/harald
- Changelog RSS feed: https://verifymcp.io/servers/k0nnect-harald/harald.xml
- Changelog JSON feed: https://verifymcp.io/servers/k0nnect-harald/harald.json
- HTML version of this page: https://verifymcp.io/servers/k0nnect-harald/harald