Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Description from the maker (original English)
Disposable Algorand wallet for AI agents. Pay x402 resources in USDC, no human provisioning.
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.
OBOLUS_NETWORK, OBOLUS_DATA_DIRYou 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": {
"obolus": {
"command": "uvx",
"args": [
"obolus",
"mcp"
],
"env": {
"OBOLUS_NETWORK": "your-value",
"OBOLUS_DATA_DIR": "your-value"
}
}
}
}Run this in your terminal
claude mcp add obolus --env OBOLUS_NETWORK=your-value --env OBOLUS_DATA_DIR=your-value -- uvx obolus mcpFill in the empty values before running it.
Paste into .cursor/mcp.json in your project, or ~/.cursor/mcp.json
{
"mcpServers": {
"obolus": {
"command": "uvx",
"args": [
"obolus",
"mcp"
],
"env": {
"OBOLUS_NETWORK": "your-value",
"OBOLUS_DATA_DIR": "your-value"
}
}
}
}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": {
"obolus": {
"type": "stdio",
"command": "uvx",
"args": [
"obolus",
"mcp"
],
"env": {
"OBOLUS_NETWORK": "your-value",
"OBOLUS_DATA_DIR": "your-value"
}
}
}
}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.
OBOLUS_NETWORK envOBOLUS_DATA_DIR envThese 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 inviti8, 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 (0.2.3) reached the registry 2 months 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, Signoff
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.
Five gates that make an agent's 'done' provable: run them, read the verdict, audit the record.
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.
Obolus
An MCP server that gives an AI agent a disposable Algorand wallet , so it can pay for x402 resources without a human provisioning anything.
The name is the coin paid to the ferryman for passage: small denomination, one purpose, spent and gone.
agent → x402_fetch(url) → 402 → sign → pay → body
x402 has over a thousand listed resources and almost no buyers — a handful of wallets account for nearly all volume, and they are scripted loops rather than agents. The rail exists; nothing can reach it. Obolus is the buyer-side piece: account setup, opt-in, signing and sweeping handled invisibly, so an agent just spends.
/plugin marketplace add inviti8/Obolus
/plugin install obolus@heavymeta
Wires up the MCP server and adds /obolus:fund-wallet, which knows the bootstrap order and why it cannot be rearranged. Nothing to install first beyond uv — the server is launched with uvx obolus mcp.
uv tool install obolus
uv tool install rather than uvx for the CLI: uvx re-resolves on every invocation, measured at 200–400 ms over an installed console script. That is nothing once per session, which is all the MCP server costs — it is why the plugin above uses uvx — but it is the wrong trade for a command called in a loop. pipx install obolus works too.
claude mcp add obolus -e OBOLUS_NETWORK=testnet -- obolus-mcp
{
"mcpServers": { "obolus": { "command": "obolus-mcp", "env": { "OBOLUS_NETWORK": "testnet" } } } }
It starts on testnet. Mainnet is never disabled — Obolus is meant to make real payments — but it is never what you get by forgetting to choose.
A new vault holds nothing, and the order is forced by the chain:
obolus vault # says which of the three steps you are on
obolus vault qr # scannable codes for the two things you can send
| # | Step | Who |
|---|---|---|
| 1 | Send ≥ 0.21 ALGO to the vault | you |
| 2 | Opt the vault into USDC (obolus vault optin) | Obolus |
| 3 | Send USDC | you |
Step 3 before step 2 fails. An Algorand account cannot receive an asset it has not opted into — the transfer is rejected outright, it does not sit pending. That is why the QR codes encode the asset id and not just the address, and why wallet_funding_info tells you which step you are on every time rather than only on error.
The QR codes carry no amount. You type that into your own wallet, where you see it before confirming.
Three tools, deliberately:
| Tool | What it does |
|---|---|
x402_fetch | Fetch a URL, paying if it answers 402. Returns the body. |
wallet_status | Balances, the active session, spending today, the caps. |
wallet_funding_info | How to put money in. For the human, not the agent. |
There is no tool for any particular merchant. Everything is reached through x402_fetch, because the moment a wallet grows first-class verbs for one seller it stops being a wallet.
The session balance, enforced by the chain. Payments come from a short-lived session account funded from your vault, never from the vault itself. A session cannot spend what it does not hold, and that is true regardless of what any code here does.
Everything else is in-process and configurable in config.toml inside your data directory:
[caps]
session_balance_micro = 5000000 # $5 per session per_call_micro = 500000 # $0.50 per call daily_micro = 2000000 # $2 a day, resets 00:00 UTC allowlist = [] # payTo addresses or hostnames; empty = any
[session] idle_timeout_seconds = 600 # close and sweep back after this
[files]
Some refusals are not caps and have no config key: paying an address we control, a non-https resource on mainnet, an asset the wallet does not hold, and any challenge that is not a well-formed x402 v2 402.
Your seed is the only way back to your money. It lives in your data directory. Pointing OBOLUS_DATA_DIR at a new location creates a *new* wallet; it does not move the old one. Back the directory up.
File access is off by default. x402_fetch can send a file to a paid endpoint and write the response back, which is how you get an image signed or a document processed. Both are disabled unless you set [files] root, and confined to that directory when you do. This is deliberate: moving bytes off your machine is not something a spend cap can bound.
Approval is per tool, not per payment. Most MCP clients ask once and remember. That means the caps above are your real spending boundary, not the prompt — see DESIGN.md §7.1, which documents what was measured rather than what was assumed.
A settled payment proves settlement and nothing else — not that the resource was correct, honest, or worth its price.
Working on mainnet. On 2026-08-13 and 08-14 an agent used Obolus over MCP to buy three resources from a live merchant with real USDC — a notarisation at $0.05, a second at $0.05, and a C2PA signature over a 13 MB TIFF at $0.15. Each payment came from a disposable session account that was derived, funded, spent and swept closed; two sessions, both reconciled, nothing stranded. All three settlements landed with fee: 0, sponsored by the facilitator, as designed.
What that demonstrates is the rail, end to end: an agent with no human in the loop held a key, answered a 402, and got the bytes. It does not say the resources were worth their price — settlement proves settlement and nothing more.
Testnet remains the default and the place to develop. Mainnet is guarded, never disabled.