WattGap for Agents

What this is

WattGap maintains a certified, per-address-class ground-truth layer for US home energy programs: utility service territory, rebate and incentive programs, program status (open, closed, launching), and eligibility screening, each fact sourced and dated. The MCP server exposes four read-only tools over that layer. It is not a chatbot wrapper around a search index; every response is validated against the same runtime contract that backs WattGap's REST API, so a tool call and a REST call return identical data for identical facts.

This is the differentiator: not the answer alone, but what stands behind it. Every response carries its own provenance -- which sources were checked, when they were last verified, and how confident the result is -- so a calling agent (and the household or institution it serves) can tell a certified fact from an estimate from a genuine unknown.

The four tools

All four tools take exactly one input: a five-digit US ZIP code. None accept or request income, address, household size, identity, account, document, payment, or consent data.

Every tool is marked read-only, non-destructive, idempotent, and open-world in its MCP annotations, and declares noauth in its security metadata -- accurate today, since no key is required to call it.

Provenance and freshness on every answer

Every successful response carries a meta.provenance block: the source list, a verified_at timestamp, a confidence tier (high, medium, or low), age in days, a stale flag, and the reason behind the confidence call. A high confidence tier means WattGap has verified ground-truth program data for that fact; medium means the value is estimated from regional signals; low means only generic fallback data was available. WattGap never upgrades a confidence tier, a ranking, or a recommendation to make an answer look better than the underlying evidence supports, and it never fills a gap with an unmarked guess -- an honest "unknown" is a product behavior, not a failure state.

Access: no-charge, anonymous, read-only

The MCP surface is open today: no API key, no registration, no OAuth handshake. Call volume is bounded by published rate limits (below), enforced per calling client from the connection itself rather than from any self-declared label -- this surface accepts no caller identifier, and nothing a caller sends changes the answer, the throughput, or the eligibility, ranking, or confidence returned. This is WattGap's no-charge tier: free to call, never worth gating truth behind. A metered commercial tier for high-volume platform integrations is a later, separately announced step; it will change rate limits and billing only, never the answer.

Rate limits (as implemented)

A request that exceeds a limit is rejected with a bounded response; the gateway never blocks indefinitely or lets one caller starve another. The gateway is stateless by design: it accepts no sessions and no request batching, so every call is independently rate-limited and independently answered.

What this will never do

Anti-gaming posture

This surface is intentionally anonymous and stateless today, which means raw call volume is not, by itself, evidence of real adoption. WattGap knows this and has already decided how it will measure adoption honestly rather than let an unauthenticated script manufacture a false signal: anonymous, no-charge discovery stays exactly as open as described above, and a separate authenticated tier (reusing WattGap's existing two-tier API-key system) will be the basis for any adoption or usage claim WattGap makes externally. That authenticated tier ships at the same activation step that lists this surface in a public registry, so the measurement exists before the number does.