get_service_statusLABSCO SUMMARY
A fresh client can mint its own key, which is unusual and deliberate.
Two of the three tools need no auth at all, so an agent can connect with no credentials, call provision_api_key, and start asking — no signup page, no approval queue. The design detail that matters more day to day is the long-poll: 20 seconds, then a pending response with both a URL for the human and one for the agent, so nothing blocks indefinitely waiting for someone to look. The README is also clear about when not to use it — deterministic, verifiable or easily reversible work should not go to a person.
What it is
A hosted human-in-the-loop service. When an agent reaches a call it should not make alone — a deploy, a delete, a judgement about taste — it posts a title, optional context and two to four labelled choices. A person is shown the request, picks one, and the agent gets back the selected id and label. The same backing API is reachable over REST and from a browser SDK.
What you get
ask_humansubmits a decision and long-polls for up to 20 seconds; if nobody has answered by then it returnspendingwith areview_urlfor the human and apoll_urlfor the agentprovision_api_keymints a token with no auth, so a fresh client can bootstrap itselfget_service_statusreports availability and escalation guidance, also with no auth- Requests carry optional context:
content_typeas text, markdown, HTML or an image, plusconfidence,consequence_of_wrong_choice,recommended_optionanddecision_reason timeout_secondsdefaults to 3600 and goes up to 86400- A
callback_urlwebhook, which must be HTTPS and host-allowlisted, signed withX-HITL-TimestampandX-HITL-Signature - Key provisioning is rate-limited to 5 keys per email
Requirements
Nothing to install — it is a remote server at https://meatspace.run/api/mcp over Streamable HTTP. get_service_status and provision_api_key need no credentials, so a client can connect cold and mint its own token; ask_human then needs that token as a Bearer header. Registered as version 0.2.0 in the MCP registry. The obvious dependency is a human: if nobody opens the review URL, the call times out and returns pending.
Setup effort
Paste a URL with a key — add the endpoint to your client
USE CASES
Gating a deploy behind a human yes
Post the state of the tests and two choices, and let a person decide before anything ships.
Breaking a tie the model is not confident about
Send the options with a confidence score and the consequence of getting it wrong attached.
Getting a judgement call on copy or design
Send markdown or an image as the content and let someone pick the version they prefer.
Bootstrapping an agent with no setup step
Connect, check the service is up, mint a key, and start asking — all in the same session.
3 tools
Listed in project docs. Your client may show a different list.
provision_api_keyask_humanBEFORE YOU INSTALL
What to know first
Tools you get
3 tools named in the project's documentation.
Requires
A client that can reach a remote MCP server over HTTP
Where it runs
The documented client connects to a hosted endpoint.
INSTALLATION
Connect your client
Add the endpoint to your client.
Server setup
https://meatspace.run/api/mcpMCP Clients
Claude Code
{
"mcpServers": {
"meatspace": {
"type": "http",
"url": "https://meatspace.run/api/mcp",
"headers": {
"Authorization": "Bearer <YOUR_API_KEY>"
}
}
}
}<YOUR_API_KEY> is a MeatSpace key. It is free and needs no signup: connect once without the headers block and call provision_api_key (it needs no auth), or POST your agent's name and an email to https://meatspace.run/api/keys. The key is shown once, so save it. get_service_status and provision_api_key work without it; ask_human needs it.