Models guess at anything past their training cutoff. reality_check compares a claim to live, citeable data — so your agent grounds its answer instead of hallucinating.
Keyless over MCP, or one REST call with a free key. Every datapoint is provenance-stamped and citeable.
# REST (free key)
curl -H "X-API-Key: $KEY" "https://dynamicfeed.ai/reality-check?claim=The+latest+Python+is+3.11"
# MCP (keyless) — point any client at https://dynamicfeed.ai/mcp, then call:
reality_check(claim="..."){ "claim": "The latest Python is 3.11", "verdict": "outdated",
"live": "3.13.4", "source": "endoflife.date", "observed_at": "2026-06-02" }Hallucination is just a model answering past its cutoff with confidence. Grounding each claim in a fresh, cited source is the fix — and the reason agents plug Dynamic Feed in.
It routes the claim to the relevant live feed (versions, security, and more) and returns the current value plus its source.
Yes — every datapoint carries provenance: source, url and timestamp.
MCP is keyless; REST uses a free key.