The sandbox: test keys
A key that starts mp_test_ is the sandbox. There is no second server and no second address: a test key calls the same API and the same MCP server, and what it makes is kept apart.
What a test key does
- It creates test polls, and among the account's polls it works on those alone.
- A test poll's run is answered by a stub model, so no AI provider is called and nothing is charged. A call that would spend needs no
max_cost_usd. - Nothing a test key makes is ever public, and none of it appears in the account holder's own list of polls.
What is the same
Everything else, so what you build against a test key works unchanged with a live one:
- The addresses:
https://mindpolling.com/api/v1andhttps://mindpolling.com/mcp. - The limits: 600 reads and 60 writes a minute per key, with
RateLimitheaders on every answer. - Validation, every error and its
code, andIdempotency-Keyon writes.
Get a test key
The account holder makes one on the API keys page, choosing test mode and the scopes it needs. An account is free and a test key needs no balance. A program cannot make a key: how keys work.
Try it
Check the key. The answer says it is a test key and lists what it can do:
curl -s https://mindpolling.com/api/v1/me -H "Authorization: Bearer mp_test_..."Then, in order: create_poll makes a test poll, run_poll_now queues a run and answers 202 with where to follow it, get_live_run follows it, and get_results reads what the stub answered. The API guide has each call with its fields.
Over MCP, add https://mindpolling.com/mcp to a client with the test key as Authorization: Bearer, and call the same four as tools.
See also
- For developers: the API, the MCP server and keys.
- How keys work: scopes, spending limits and what is logged.