Verifiziert kostenpflichtige AI Agent Services mithilfe von signierten Belegen und Echtzeit-Messungen, um Fehlzahlungen zu verhindern.
Auch bekannt als botsrated/botsrated, botsrated, @botsrated/x402-preflight
- Stars
- 0
- Tools
- 12
- Downloads, 30 Tage
- 2.227
- Neueste Version
- 0.4.0-beta
- Letzter Commit
- 12. Nov. 2026
- Lizenz
- MIT
Was ist BotsRated MCP?
BotsRated ist ein MCP server. Verifiziert kostenpflichtige AI Agent Services mithilfe von signierten Belegen und Echtzeit-Messungen, um Fehlzahlungen zu verhindern. Er stellt 12 Tools bereit, etwa preflight, check_mcp_setup, watch_free.
Zusammenfassung generiert aus dem README des Projekts
Über BotsRated MCP server
BotsRated bietet eine Verifizierungsschicht für Entwickler, die kostenpflichtige AI Agent Services nutzen. Es fungiert als remote MCP server, der zeitgestempelte, durch Belege abgesicherte Messungen von x402 Agent Services auf dem Base-Netzwerk veröffentlicht. Durch die Nutzung von Tools wie preflight und check können Entwickler auf faktische Daten zu Service-Latenz, angebotenen gegenüber gelisteten Preisen und validen 402-Challenges zugreifen, um Fehlzahlungen zu vermeiden.
Dieser Service ist für alle gedacht, die Coding Agents wie Cursor, Claude Code oder Cline betreiben, sowie für Entwickler, die Frameworks wie LangChain, LlamaIndex oder das OpenAI Agents SDK nutzen. Er enthält ein optionales Guard-Paket, das Zahlungsanfragen mit Echtzeit-Messungen vergleicht, um Transaktionen basierend auf benutzerdefinierten Regeln zu warnen oder zu blockieren. Der Server läuft über streamable HTTP und erfordert keine lokale Installation, nutzt jedoch nach Überschreiten der kostenlosen täglichen Quoten x402 USDC-Zahlungen auf Base für bestimmte kostenpflichtige Tools.
Was du damit machen kannst
- 1Service-Preise vor der Zahlung verifizierenEin AI Agent kann das preflight-Tool nutzen, um den gelisteten Preis eines Verkäufers mit dem tatsächlich während eines Probes beobachteten Betrag zu vergleichen.
- 2Zahlungshistorie für einen Host auditierenDu kannst eine vollständige Historie signierter Belege abrufen, um die vergangene Performance und Abrechnungsgenauigkeit eines spezifischen Service-Hosts zu überprüfen.
- 3Änderungen des Service-Status überwachenEntwickler können Alerts einrichten, um Benachrichtigungen zu erhalten, wenn ein kritisches Tool seine payTo-Adresse oder den angebotenen Preis ändert.
Geschrieben aus dem README des Projekts. Jeder Use Case wurde gegen ein Zitat daraus geprüft.
So installierst du BotsRated MCP
Formate geprüft am 10. Nov. 2026BotsRated MCP in Claude Code installieren
$ claude mcp add botsrated -- npx -y @botsrated/x402-preflightBotsRated MCP in Cursor installieren
~/.cursor/mcp.json
{
"mcpServers": {
"botsrated": {
"command": "npx",
"args": [
"-y",
"@botsrated/x402-preflight"
]
}
}
}BotsRated MCP in VS Code installieren
.vscode/mcp.json
{
"servers": {
"botsrated": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@botsrated/x402-preflight"
]
}
}
}BotsRated MCP in Claude Desktop installieren
claude_desktop_config.json
{
"mcpServers": {
"botsrated": {
"command": "npx",
"args": [
"-y",
"@botsrated/x402-preflight"
]
}
}
}BotsRated MCP in Codex installieren
~/.codex/config.toml
[mcp_servers.botsrated] command = "npx" args = ["-y", "@botsrated/x402-preflight"]
BotsRated MCP in Gemini CLI installieren
~/.gemini/settings.json
{
"mcpServers": {
"botsrated": {
"command": "npx",
"args": [
"-y",
"@botsrated/x402-preflight"
]
}
}
}BotsRated MCP tools 12
Erkannt aus dem READMEpreflightcheck_mcp_setupwatch_freewhat_changedfindcoveragesamplecheckreportbestseller_self_checkseller_announce
| preflight | — | |
| check_mcp_setup | — | |
| watch_free | — | |
| what_changed | — | |
| find | — | |
| coverage | — | |
| sample | — | |
| check | — | |
| report | — | |
| best | — | |
| seller_self_check | — | |
| seller_announce | — |
Automatisierte Checks
Geluften am 12. Nov. 2026Vollständiger Bericht →Unsere automatisierten Checks bestanden. Das ist kein Security-Audit: Lies den Code und gib dem Server nur den minimal notwendigen Zugriff. Wie die Checks funktionieren
- Source Repository verknüpft und erreichbarBestanden
- Open Source Lizenz (OSI)Bestanden
- Commit oder Release im letzten JahrBestanden
- Package-Name ähnelt keinem populären PackageBestanden
- Remote Endpoint nutzt HTTPSBestanden
- Destruktive Tools sind markiertBestanden
- Keine versteckten Anweisungen an das ModellBestanden
- Kein kritischer Hinweis im Paket oder seinen AbhängigkeitenNicht geprüft
README
Auf GitHub ansehenBotsRated: MCP server
Signed checks of paid AI services, before your agent pays.
BotsRated measures paid AI-agent services with signed receipts. Paid BotPulse plans are retired; the feed stays free. These measurements do not claim coverage of every delivery outcome.
BotsRated publishes timestamped, receipt-backed measurements of paid x402 agent services on Base and Solana:
- unpaid probes: status, whether the 402 challenge is valid, latency, quoted vs listed price, and payTo;
- paid test-buys, each with its on-chain tx hash.
It reports facts and signed receipts only.
This repository documents the remote MCP server (nothing to install or run for it) and the optional Guard packages below.
Guard: check a payment before your agent signs it
Guard pairs with the x402 client your agent already pays with. It does not replace your agent runtime. Free. No key for local attach. It compares each payment request with BotsRated measurements and warns or blocks by your own rules. Facts and receipt links, not grades. Guard does not hold funds.
- Coding agents (Cursor, Claude Code, Cline, Codex): follow
llms-install.md→ When scaffolding x402 pay. The same recipe ships as a Cursor rule (.cursor/rules/botsrated-guard-attach.mdc) and an agent skill (docs/agent-skills/botsrated-guard-attach/SKILL.md). If a copy drifts,llms-install.mdwins. Attaching Guard before the first paid request is a required step of that recipe, not an option. - Routers and wallets: call Guard before pay with
llms-install.md→ Router or wallet: call Guard before pay (POST /v1/guard/checkor MCPguard_check, spec indocs/guard/guard-check.md). This route and the MCPguard_checktool are not live on botsrated.com yet. Optional, read-only, does not pay. - How attaches are counted: outside Guard attaches that come back week over week, excluding internal calls. Not npm or PyPI downloads. See
llms-install.md→ How we count outside attaches. No public figure yet. - Default path: hook → measured seller facts → your policy pass / fail receipt → Desk / alerts handoff. Canon:
docs/guard/default-middleware.md. - Per stack: integrations/README.md · live docs https://botsrated.com/docs/guard
| Stack | Install | Attach |
|---|---|---|
| LangChain | pip install langchain-botsrated |
README |
| OpenAI Agents SDK | pip install botsrated-openai-agents |
README |
| LangGraph | pip install langgraph-botsrated |
README |
| Llama Stack | pip install llama-stack-botsrated |
README |
| LlamaIndex | pip install llama-index-tools-botsrated |
README |
| Microsoft Agent Framework | pip install agent-framework-botsrated |
README |
| Vercel AI SDK | npm install @botsrated/ai-sdk |
README |
| Any Python x402 client | pip install "botsrated[x402]" |
README |
| Any TypeScript x402 / MCP client | npm install @botsrated/x402-preflight |
README |
Versions and the stacks not listed here (Google ADK, CrewAI, Pydantic AI, Mastra) are in integrations/README.md.
| MCP endpoint | https://botsrated.com/mcp (Streamable HTTP, JSON responses, no auth) |
| Official MCP Registry | com.botsrated/botsrated |
| Website | https://botsrated.com |
| OpenAPI | https://botsrated.com/openapi.json |
| A2A agent card | https://botsrated.com/.well-known/agent-card.json |
| llms.txt | https://botsrated.com/llms.txt |
| Contact | botsrated@gmail.com |
Tools
Live MCP tools/list at https://botsrated.com/mcp has 12 tools:
| Tool | Price | What it returns |
|---|---|---|
preflight |
free, no key, rate limits | Facts bundle for an x402 seller or remote MCP server (host or URL): latest unpaid probe, freshness, directory listing, receipt link. Measurements and a receipt link. See below |
check_mcp_setup |
free | Check the MCP servers in a config or a list of hosts (up to 50 hosts or an mcp.json). Same report as POST /v1/dependency-watch/report |
watch_free |
free | Email when a covered host's measured facts change (1 host, at most once per UTC day, 30 days, up to 3 active watches per client). Same watch as POST /v1/alerts/subscribe |
what_changed |
free | Measured changes for a free watch. Pass the handle from watch_free |
find |
free (uses 1 of the 25 free checks per client IP per UTC day) | Top matching x402 sellers and remote MCP servers for a need (q, e.g. weather), each with a one-line factual reason and its page URL. Same filters as best, without the full ranking. HTTP: GET https://botsrated.com/v1/find?q= |
coverage |
free | How many x402 sellers and MCP servers are covered, from which sources, and when the last probe ran. Live counts: GET https://botsrated.com/v1/coverage |
sample |
free | A real sample of a paid answer (check, report or best) for a fixed example |
check |
$0.01 (x402, USDC on Base); first 25 per client IP per UTC day free on this tool and on HTTP GET /v1/check (one shared quota) |
Measured facts and the last 5 receipts for one host. Paid at GET https://botsrated.com/v1/check?host= |
report |
$0.002 (x402); paid on MCP from the first call. The first 3 reports per client IP per UTC day are free only on HTTP GET /v1/report/{host} |
Full receipt history for one host. Paid at GET https://botsrated.com/v1/report/{host} |
best |
$0.01 (x402); paid on MCP from the first call; no free quota | Top 3 measured providers for a task (sort=reliability|price|latency). Paid at GET https://botsrated.com/v1/best?task= |
seller_self_check |
free | Stored measurements for a host anyone can read: listing versus quoted price and recent unpaid probes |
seller_announce |
free (verified owner) | Verified owner posts a planned price and/or payTo |
Facts bundle (preflight and check)
preflight (the MCP tool and GET /v1/preflight?url=<host>) returns measurements and receipt links. You set your own
thresholds. The response is measurements and a receipt link.
| Field | What you get back |
|---|---|
found, host |
Whether this host is in the probe record, and the host name |
latest_probe |
Latest unpaid probe: ts, status, valid_402, latency_ms, observed amount (live_amount), observed payTo (live_payTo), live_network |
latest_probe.receipt_url |
Link to the signed receipt for that probe (https://botsrated.com/receipt/...). The receipt is the record |
freshness |
last_probe_at, age_seconds, stale (older than 24 h) |
listing |
Directory price and payTo. Directory data is not a probe observation |
recent_window |
Probe count, how many returned a valid 402, latency. A count in a window |
note |
Facts only |
check (paid, after the free daily quota) adds measured facts for one host and the last receipts. Those are still
measurements and receipt links.
The payment guard in @botsrated/x402-preflight and botsrated keeps a smaller local bundle from this response
(covered, host, knownPayTos / known_pay_tos, listingPayTo / listing_pay_to, observedNetwork /
observed_network, observedAmount / observed_amount, stale, receiptUrl / receipt_url). Pick a named preset
(warn, carefulBuyer, or dry-run-only) there. The guard does not invent a summary number from these fields.
Paid MCP tools: an unpaid call returns an x402 v2 PaymentRequired tool error (isError, structuredContent with accepts); an x402-aware MCP client (e.g. @x402/mcp) retries with the payment in _meta["x402/payment"]. The same data is on the HTTP routes, payable by any x402 client (e.g. @x402/fetch, Coinbase Agentic Wallet). All three paid routes are also listed in the Coinbase x402 Bazaar.
Free daily quota: the first 25 checks per client IP (IPv6 per /48) per UTC day are free on the MCP check tool and on HTTP GET /v1/check (one shared quota). The first 3 reports per client IP per UTC day are free only on HTTP GET /v1/report/{host}. MCP report and best are paid from the first call. /v1/find spends one check from the shared check quota. /v1/best has no free quota. After the free quota the normal x402 402 applies. Preflight stays free, no key, rate limits. No accounts, prepaid balances or stored credit. The badge is free and earned, never sold. Details: https://botsrated.com/pricing
Free HTTP endpoints and pages
GET https://botsrated.com/v1/find?q=<need> |
Top matches for a need with a reason (uses 1 of the 25 free checks per client IP per UTC day, shared with /v1/check) |
GET https://botsrated.com/v1/pulse |
Live totals (cached 60 s): services watched, working in the last 24 h, down or failed in 7 days, checks run in 7 days. Never names a seller |
GET https://botsrated.com/v1/savings-rates |
Measured rates (offline, payTo changes, price vs listing) with sample size, dates and source |
| https://botsrated.com/sellers | Browse and filter every covered seller and MCP server |
https://botsrated.com/s/<host> |
One page per seller or MCP server: status, 30-day chart, price, payTo, test purchases, receipts, alternatives |
| https://botsrated.com/savings | What we check for, with the rates we measured |
| https://botsrated.com/leaderboard | Data table of the top 300 covered x402 sellers |
Plans (seller alerts and data plans)
One x402 USDC payment on Base buys one 30-day period. There is no balance and no auto-renewal. Cancelling stops delivery. The dataset stays free.
Pro: watch up to 50 sellers or tools; alert to webhook, Slack, Discord or email. $19 / month by card + tax, or $19 / 30 days USDC. Tools you rely on (for example OpenAI, Anthropic, Gemini, Pinecone, Supabase) use those same 50 slots. No separate plan.
| Plan | Price per 30 days | What you get |
|---|---|---|
alerts-free |
$0 | Watch 1 seller. Daily digest: one email per UTC day when its facts change. Slack, Discord, and webhook need a paid plan (paid_channel_required). Up to 3 free alerts per client IP, 1,000 in total |
alerts-50 (Pro) |
$19 | Up to 50 sellers or tools; a signed webhook within one probe cycle of a change (payTo, quoted price, valid 402, listing, freshness) |
desk-pro (Desk Pro; API name data-pro) |
$199 | Bearer token for /v1/check, /v1/report, /v1/best without per-call payment (600 requests/min), the change-event export (GET /v1/export/changes.ndjson?host=), and the team workspace |
| Data licensing | No public price | Volume or bulk delivery; support terms agreed per contract: botsrated@gmail.com |
data-starter is not on sale; a token already issued keeps working until it expires. Card plans: Desk Team $99/month, Desk Pro $199/month, Pilot $149 for 6 weeks (paid once). Current prices: https://botsrated.com/pricing
Paid BotPulse plans are retired; the feed stays free. Plans already bought keep working until their 30-day period ends. Free reads stay at https://botsrated.com/v1/botpulse/latest and https://botsrated.com/v1/botpulse/feed.json. Alerts on tools use Pro's 50 slots.
Subscribe with POST https://botsrated.com/v1/alerts/subscribe (free) or POST https://botsrated.com/v1/subscribe/{plan} (x402). The body and the webhook are checked before any 402. The webhook must answer the verification POST with the challenge; it must be https on a public DNS name, and redirects are not followed. Free alerts are email only: one confirmation link first, then plain-text alerts from alerts@botsrated.com. Slack, Discord, and webhook delivery need a paid plan (paid_channel_required on the free plan). Paid alert plans can use a signed webhook (Ed25519, botsrated-prober-1, see below, over {"m":"<timestamp>.<raw body>"}), a Slack incoming webhook or Discord channel webhook (plain text, not signed), or the same email path. Manage a subscription with GET /v1/subscription and POST /v1/subscription/{cancel,update} plus Authorization: Bearer <token>. Full details: https://botsrated.com/v1/plans
Install
Claude Code:
claude mcp add --transport http botsrated https://botsrated.com/mcp
Cursor / Windsurf / Cline / any client that takes a JSON config:
{
"mcpServers": {
"botsrated": { "type": "streamableHttp", "url": "https://botsrated.com/mcp" }
}
}
VS Code (.vscode/mcp.json):
{ "servers": { "botsrated": { "type": "http", "url": "https://botsrated.com/mcp" } } }
Changelog
See CHANGELOG.md. Catalog listing copy (not submitted from this file): SUBMISSIONS.md.
Security
Report security issues privately: see SECURITY.md (botsrated@gmail.com, subject "Security"). Contributing: CONTRIBUTING.md.
Methodology, replies
- Methodology: https://botsrated.com/methodology
- A seller can send a reply about a measurement for free: https://botsrated.com/disputes (
POST /v1/dispute). The measurement is re-checked and the reply is published next to the data. A measurement is a signed record, not a finding about the seller. - Dataset signature: https://botsrated.com/dataset.sig.json. Key set: https://botsrated.com/.well-known/botsrated-key
- Receipt format v0.2.1 (draft;
mcp_probereceipts for remote MCP servers; v0.2.1 updates reference schedules and MCP retention) and verifier:spec/SPEC.md.node spec/verify.mjs '<receipt.json|url>'needsverify.mjs,verify-core.mjs, andverify-node.mjsin one directory.verify.mjsalone does not run. - Free: preflight, then the receipt link, then the signature check. x402 preflight puts it on
latest_probe.receipt_url. MCP preflight puts it onservers[].latest_receipt_url(keep?mcp=&day=&s=). CLI, single quotes:node spec/verify.mjs '<receipt_url>'. The pasted URL is a 20-hex id plus allowlistedhost,mcp,day, andsonly. A matching signature checks that record only. It is not a rating of the seller. The same check is importable asverifyReceiptfrom@botsrated/x402-preflight/verify-receipt(Node, offline; statussignature valid,signature invalid,key unknown, orredacted copy). A copy withwithheld_fieldsorwithheld_noteisredacted copy: not offline-verifiable, and notsignature invalid. Check the stored record withGET /v1/verify.spec/receipt-page.mjsis that same check forAccept: text/html. It is live on the host (Content-Security-Policyheader, and/spec/verify-browser.mjsplus/spec/verify-core.mjs).spec/courthouse/is the same check as a folder you serve yourself.
Signing keys (Ed25519)
Raw 32-byte public keys, base64. Fingerprint = sha256 of the raw key bytes. A signature is valid for an object whose own timestamp
(ts, or generated_at for snapshots) lies in [valid from, valid until). The live key set at /.well-known/botsrated-key is authoritative; this table is an out-of-band copy to compare against.
| key_id | Status | Valid from (UTC) | Valid until (UTC) | Signs | Public key | sha256 fingerprint |
|---|---|---|---|---|---|---|
botsrated-prod-1 |
active | 2026-10-23T15:00:00Z | 2027-10-23T00:00:00Z | dataset snapshots, paid test-buy receipts | oTK25Dq3wU2+apPYddiv4T+8is2s3hL1k6ozjMAbEWQ= |
ac58fa77b9f6a26805efb65d331d6d77da3dfbdbbe981cc7510692852a50e74a |
botsrated-prober-1 |
active | 2026-10-23T12:00:00Z | (none) | cron probe receipts, daily aggregates, archive manifests | G5aD8mz2rkbpLPQuEC2t50AaJpMjGcflTXuRmeJETsE= |
805e75f8795ff38c4805653962ba75991fc8bdcaef3f80441e0775e405a3a628 |
botsrated-staging-1 |
retired (verify only) | 2026-10-22T00:00:00Z | 2026-10-23T15:00:00Z | dataset snapshots generated before 2026-10-23T15:00:00Z | OfGrLtyJYCC1K9jjhvq0Zduwgqq68DvzR1StOo5ZpO0= |
71db9690ef765ce6b923d63fc71e1a65474e87ee1073532eafb9099b1e6f0288 |
Payments are USDC on Base mainnet.
BotsRated MCP Fragen
Wie füge ich BotsRated MCP zu Claude Code hinzu?
Führe claude mcp add botsrated -- npx -y @botsrated/x402-preflight in deinem Terminal aus. Füge --scope user hinzu, damit der Server in jedem Projekt verfügbar ist.
Wie füge ich BotsRated MCP zu Cursor hinzu?
Öffne ~/.cursor/mcp.json und füge den Server unter mcpServers hinzu, mit der Cursor-Config aus dem Installationsabschnitt oben.
Welche Tools bietet BotsRated MCP?
Er stellt 12 Tools bereit, darunter preflight, check_mcp_setup, watch_free, what_changed, find.
Wie kann ich Guard nutzen, um die Zahlungen meines Agents zu schützen?
Guard vergleicht Zahlungsanfragen mit BotsRated-Messungen, um Zahlungen basierend auf deinen eigenen Regeln zu warnen oder zu blockieren, ohne dabei Gelder einzubehalten.
Gibt es eine kostenlose Stufe für die Überprüfung von Services?
Ja, die ersten 25 Checks und 3 Reports pro Client-IP und UTC-Tag sind kostenlos, und das Preflight-Tool ist mit Rate Limits ebenfalls gratis.