Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Description from the maker (original English)
helps coding agents work out the bluetooth le and classic protocol of gadgets you own
These three are never merged into one score: opinions, measurements and popularity are different things.
We have not written a plain-language summary for this one yet. Below are the facts we read from the registry.
Pick the AI app you use, then copy the settings.
You need
Which app do you use?
Pick your app above. We remember it, so next time the button says “Install in …”.
Open Claude Desktop → Settings → Developer → Edit Config, and paste this into claude_desktop_config.json
{
"mcpServers": {
"harald": {
"command": "uvx",
"args": [
"harald",
"mcp"
]
}
}
}Run this in your terminal
claude mcp add harald -- uvx harald mcpPaste into .cursor/mcp.json in your project, or ~/.cursor/mcp.json
{
"mcpServers": {
"harald": {
"command": "uvx",
"args": [
"harald",
"mcp"
]
}
}
}Your browser asks to open the app, then the app shows the settings and asks you to confirm. Key values are left empty for you to fill in.
Paste into .vscode/mcp.json in your project
{
"servers": {
"harald": {
"type": "stdio",
"command": "uvx",
"args": [
"harald",
"mcp"
]
}
}
}Your browser asks to open the app, then the app shows the settings and asks you to confirm. Key values are left empty for you to fill in.
This one runs on your computer, and ChatGPT can only connect to servers on the internet. Pick another app.
Your choice stays on this device in a cookie; no account needed.
These settings are generated from the maker’s registry entry. If the README says otherwise, follow the README.
Runs on your computer; the tool list appears in your AI app after you install it.
We check information anyone can see; we have not read the code. Checks that do not apply are left out of the count. · Checked Nov 7, 2026
The registry name was verified with the GitHub account k0nnect, which owns the repo
We have not read this repo from GitHub yet
We have not read this repo from GitHub yet
This one runs on your computer; there is no internet endpoint
Runs on your computer and uses local settings instead
Runs on your computer; we can only read tool lists from internet endpoints
Compared with every active server in this category in the registry. Stars count only when the publisher owns the repo.
It passed 1 of 1 applicable basic checks. We have not read the code. Before adding a key or token, see what access the server asks for and use a key with only the permissions you need.
The registry does not list a required key, though some servers ask you to sign in on first use.
It runs on your computer. Your AI app starts it when needed, and data does not pass through the maker’s servers (unless it calls an online service itself).
The GitHub repo last changed –, and the latest version (0.2.2) reached the registry 4 days ago.
Apps that support MCP, such as Claude Desktop, Claude Code, Cursor and VS Code. You also need Python (uv) on your computer.
In the same category: Spec Driven Development, Remote Desktop Commander, AvaHR
Servers in the same category from other makers, sorted by stars.
Reading a card:★ GitHub starsiThree pillars: reviews · checks · usageiUpdated = latest version published to the registryi
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
Read hiring data and create inactive assessment and questionnaire drafts for human review.
Copied from the maker’s repo as written. PopNx MCP has not reviewed or endorsed it. HTML, images and scripts are removed. Links lead off this site.
install / how it works / devices / bluetooth classic / add your device
fig. 1 · claude + harald find a sony wh-1000xm5's control channel over bluetooth classic, wait for you to arm it, then turn noise cancelling on. real hardware, sped up.
harald is an mcp server, a cli and a playbook for coding agents. it gives your agent hands and ears for bluetooth le and classic, so it can work out the protocol of a gadget you own. then you control that gadget from your own code instead of the vendor's app.
| *listen* | *decode* | *speak* |
|---|---|---|
| harald reads what the vendor app says to the gadget, from your phone's bluetooth log. | your agent works out the language, keys and checksums included, by diffing payloads, testing keys and searching the vendor's android app. | you get the protocol written up and a standalone driver script with its own small cli. no vendor app. |
you need uv. then:
uv tool install harald
that puts harald on your path for the commands below. if your shell can't find it, run uv tool update-shell. to try harald without installing it, run uvx harald --help: uvx fetches harald and runs it once. tested on mac; linux is experimental, see below.
the first time harald scans, macos asks whether your terminal app may use bluetooth. allow it. if you missed the prompt, turn it on in system settings › privacy & security › bluetooth.
two optional tools, for the steps that need them:
brew install jadx lets harald read a vendor's android app.brew install libimobiledevice records your iphone's bluetooth log from the terminal.harald also works without an agent:
harald scan list bluetooth le devices nearby
harald connect list a device's services and characteristics harald read read a characteristic once harald write write bytes, once you have armed the device harald capture open show a phone capture as a timeline harald apk search a vendor android app for uuids and keys harald devices list the device library harald drive run a library device's driver
harald --help lists the rest, and bluetooth classic has the commands for classic gadgets.
linux (experimental)
harald works on linux too, for bluetooth le and classic, through bluez. you need bluez 5.50 or newer with bluetoothd running, and your user in the bluetooth group: log out and back in after adding it. harald needs no root, and there is no permission prompt as on macos, but it can't reach bluetooth from inside a sandbox such as snap or flatpak. the install above works as it is, on uv's own python too. if harald says bluetooth is off, run bluetoothctl power on, with rfkill unblock bluetooth first if it is blocked, and check bluetoothd with systemctl status bluetooth.
pair a classic gadget in your own terminal, in this order:
bluetoothctl pairable on. without it, bluez makes a bond that is gone after the next disconnect.bluetoothctl scan on, until the gadget shows up.bluetoothctl pair .bluetoothctl trust .if harald says the gadget has forgotten the pairing, run bluetoothctl remove and pair it again.
linux is experimental: tested against an emulated gadget, not yet on real hardware. if you run it on linux, tell us how it went in an issue.
| device | what harald found | status |
|---|---|---|
| sony wh-1000xm5 headphones | noise cancelling, ambient and battery over bluetooth classic | decoded |
| your device | open a pull request with its device.toml, protocol.md and driver.py | wanted |
each decoded device lives in devices/ with its protocol written up and a standalone driver. harald devices lists them, and harald drive runs one, for example harald drive on.
add the mcp server to your agent, then install the playbook: the step-by-step method your agent follows to crack a gadget.
claude code, in your project folder:
claude mcp add harald -- uvx harald mcp
harald playbook install claude-code --dest .claude/skills/harald
use --dest ~/.claude/skills/harald to have the skill in every project.
codex
add this to ~/.codex/config.toml, or run codex mcp add harald -- uvx harald mcp:
[mcp_servers.harald]
command = "uvx" args = ["harald", "mcp"]
then, in your project folder:
harald playbook install codex
this writes AGENTS.md. if your project already has one, harald leaves it alone; use --dest ~/.codex to install it for every project instead.
cursor
add this to .cursor/mcp.json in your project, or to ~/.cursor/mcp.json for every project:
{
"mcpServers": { "harald": { "command": "uvx", "args": ["harald", "mcp"] } } }
cursor has no playbook; it uses harald's tools directly.
then ask your agent to work out your gadget with harald. it will ask you for two things only you can do.
record your phone talking to the gadget. a computer cannot overhear phone-to-gadget traffic without extra hardware, so the log comes from your own iphone. a bluetooth classic gadget's log comes from an android phone instead; see bluetooth classic.
idevicebtlogger capture.pklg, which records until you press ctrl-c..pklg file and give your agent its path. to look at it yourself: harald capture open capture.pklg.if packetlogger records nothing, remove the logging profile and install it again.
get the vendor's android app. if the commands look encrypted, your agent asks for the apk of the vendor app you already use. you don't need an android phone, and harald never downloads apps: download it yourself from apkmirror, which verifies each app's signature, or with apkeep (apkeep -a .). the package name is the id= part of the app's play store address.
some gadgets, often headphones and speakers, are controlled over bluetooth classic instead of bluetooth le: the vendor app talks to an rfcomm channel, a plain byte stream. harald reaches those too, on a mac or linux; bluetooth le works as before.
pair the gadget first: on a mac in system settings > bluetooth, keeping it connected there; on linux with bluetoothctl, as in install. harald doesn't search for or pair classic gadgets, and harald scan only sees bluetooth le. classic addresses look like aa:bb:cc:11:22:33.
harald classic devices list the gadgets paired with this computer
harald classic services list a gadget's rfcomm services harald classic send send messages, once you have armed it harald classic listen show what a channel says, sending nothing harald capture open --channel n show one rfcomm channel of a capture
is a service uuid or channel number from harald classic services; add --refresh there when the list looks out of date. a mac reads that list from its own cache; on linux harald asks the gadget itself, so the gadget has to be on and in range. harald classic send takes one or more messages and sends them in order, listening for replies after each one (--listen s, 2 seconds by default). if the grant runs out or the link fails partway, it prints what it sent and received, says why on stderr and exits 1. harald classic listen sends nothing and listens for --seconds s (5 by default). either listens for at most 30 seconds at a time.
like harald write, harald classic send needs harald arm first, and every message spends one write from the grant, a query too. harald never opens the channels of protected profiles such as hands-free or file transfer; see safety. your agent gets the same four as mcp tools: classic_devices, classic_services, classic_send and classic_listen. none of them arms.
classic captures come from android. an iphone capture of a classic gadget shows only apple's iap2 wrapper, which harald can't unwrap yet, so record the vendor app from an android phone's bluetooth log:
adb bugreport. its zip holds btsnoop_hci.log; give your agent that file's path, or look at it yourself with harald capture open btsnoop_hci.log.harald apk flags the rfcomm sockets and uuids in a vendor's android app as well as its bluetooth le calls.
harald is for gadgets you own: interoperability and research. don't use it on anyone else's devices.
harald arm <address> --writes n --minutes m
that grants a budget of n writes for m minutes (50 and 30 if you leave the flags out), and each write spends one. it shows the grant and waits for you to type arm, and it refuses to run without a terminal. harald status shows what is armed and harald disarm revokes it. there is no mcp tool that arms.
harald arm stops casual or accidental arming by an agent, but it is not a sandbox. grants live in a file, ~/.harald/arm.json, and any agent that can run arbitrary commands or write files on your machine could get around the gate. what keeps a cooperaOnly the beginning is shown; read the rest in the repo