For AI Agents
The same facts as the “For AI Agents” section of /llms.txt, in page form. If you are an agent, read that one — it is plain text and has no markup to parse around.
You can use Meta Council directly. It is not a website you have to drive through a browser: there is one hosted MCP endpoint, and everything below is reachable from it with a bearer credential.
This page is the reference. If you want the procedure instead, the Agent Quickstart walks the same ground once and in order, from no account to a finished result.
What it gives you
Put a question or decision to a panel of specialized expert agents and get back a synthesis that preserves every panelist's reasoning, confidence, dissents and risks instead of collapsing them into a single voice. Start and advance multi-step workflows that pause at human checkpoints. And work an owner-scoped operational surface — tickets, boards, epics and their dependency graph, Sales and outreach records, Invoicing, Marketing, Consulting and Accounting workspaces, plus LOCUS behavioral-health level-of-care scoring and Legal source/citation records.
This page deliberately does not list the catalogue or its size. Call
tools/list against the endpoint and read what your own credential
actually reaches, rather than trusting a number written on a web page that
goes stale the moment a tool ships.
Connect
Nothing to install:
claude mcp add --transport http meta-council https://meta-council.com/mcp --header "Authorization: Bearer mc_YOUR_KEY"Getting a key is a human step
You cannot mint your own credential today. A person signs up free, opens
Settings > Developer API, and creates a key; it carries the
mc_ prefix and is sent as
Authorization: Bearer mc_....
Ask your human to do this. There is no self-serve path for an agent, and no amount of retrying will produce one.
First three calls to try
list_panels— takes no arguments and needs no credential. Returns every expert panel with its description and agent count. Start here: it is the cheapest way to confirm you are talking to the right endpoint.run_council—{"query": "<your question>"}. Optional:panel(a slug from step 1, or"auto"to let the platform choose),model, andwait_seconds(0–90). By default it returns a session id immediately instead of blocking, so treat it as asynchronous.get_session—{"session_id": "<the id from step 2>"}. Returns the finished result: each agent's recommendation, confidence and full reasoning, plus the synthesis, consensus points, dissenting views and risk matrix.
What it costs you in quota
Limits are resolved from the tier of the account that minted the key, not from the key itself — so a key created by a free account gets that account's free allowance:
| Limit | Free account |
|---|---|
| Council runs per hour | 3 |
| Council runs per day | 5 |
| Concurrent runs | 1 |
| Daily token budget | shared across those runs |
Paid tiers raise all four. Exceeding any of them is an HTTP 429 whose
message names the limit you hit and when it resets, and the hourly and daily
refusals carry a Retry-After header — honor it rather than
retrying blind.
Honest limits, stated once
- MCP OAuth is not implemented. Bearer keys are the only way in.
- The hosted MCP surface and the
meta-council-mcpstdio package are deliberately not identical. Every tool named on this page is the hosted one. Do not assume a tool you found in one surface exists in the other. - Do not send PHI or ePHI. This is an absolute limit, not a preference. Our Trust & Security page lists what we do and do not have, control by control — including that Meta Council is not HIPAA-ready.
- You will not find usage statistics, customer counts or testimonials anywhere on this page, because there is nothing honest to put there yet. The platform is built in the open, and the production deployment we point at is the LOCUS level-of-care determination surface.
Related: /llms.txt (the same facts as plain text, plus the full API and MCP reference) · Trust & Security (what we do and don't have) · Privacy Policy · Terms of Service