{"name": "Sampling Quota Guard", "provider": "TextForAgents LLC", "support": "support@textforagents.com", "slug": "sampling-quota-guard", "what": "Meter the reverse direction: an MCP server can call back into your client's model via sampling/createMessage and spend your tokens on its work. This guard keeps a per-server sampling ledger (requests, tokens, model lobbied for, includeContext asked/effective), enforces a fail-closed per-server allowance (requests / tokens / per-request cap \u2014 denied requests never consume allowance), clamps context scope to none unless you pre-approved that server, and scores model lobbying into a pre-connect verdict (ALLOW / CAUTION / DENY / INSUFFICIENT_DATA) an agent can query before it connects.", "limits": "Honest limits: the lab census found only the corpus reference everything server samples today \u2014 the verdict-API shape is the near-term product, not a large installed metering base. Persona separation proves separability of the lobbying signal, not field accuracy, so thresholds and weights ship as operator-tunable configuration. Premium-hint matching is substring-based: an exotic model name containing a premium substring can over-score lobbying by at most its weight (default 0.35) \u2014 a verdict input only, never an allowance input, so the worst case is a CAUTION. Per-client hook check: the guard is a pure function over the raw JSON-RPC sampling request + operator policy, callable from any client sampling-handler hook without forking the client (see docs). Verdict profiles are per-operator only in this build \u2014 no cross-fleet aggregation is enabled or claimed.", "prices": {"pre_connect_verdict_usd": 0.005, "per_request_metering_usd": 0.002, "operator_flat_usd_per_month": 9, "operator_servers_max": 100, "operator_metered_included_per_month": 9000, "overage_usd_per_request": 0.002}, "trial": "First 100 pre-connect verdicts free, one time per operator account, enforced per account AND per payer wallet; a second account for the same operator does not re-grant. After the grant, the tier/API price applies to every verdict/request. There is no perpetual free tier.", "how_to_start": "POST /register {account_name, payer_wallet} -> per-customer token (shown once, SHA-256 at rest). POST /meter (Bearer) {server, request} meters one sampling/createMessage request against your per-server ledger + allowance + scope clamp ($0.002/request x402, or covered by the Operator flat tier). POST /verdict (Bearer) {server} returns the pre-connect verdict for one server from your own measured profile \u2014 the first 100 verdicts are the one-time trial, then $0.005/verdict x402 (an active flat tier covers verdicts). POST /subscribe (Bearer + X-PAYMENT $9) for the Operator flat tier. The same operations are MCP tools at POST /mcp. The embeddable guard package (lane d) ships the identical engine for client-side use. Verify any receipt at POST /receipts/verify.", "pay_to": "0x7a482F3B58780519880a6A13F983E958f2Ed6A61", "network": "base (eip155:8453)", "asset": "USDC", "signing_keys": "/.well-known/signing-keys"}