Confirme
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Confirmecheck if example.com is live and give me a signed receipt"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Confirme — did it really happen?
Independent, Ed25519-signed receipts for AI agents. You say "done"; Confirme goes and looks, then signs what it saw. A third party that belongs to no AI vendor. By Synapse, Québec, Canada.
Human page: https://www.synapseconnexion.com/ia/ (FR) · https://www.synapseconnexion.com/ia/en/ (EN)
Docs written for AIs (Markdown, token sizes announced): https://www.synapseconnexion.com/ia/index.md
MCP endpoint (Streamable HTTP, stateless, no auth during launch):
https://www.synapseconnexion.com/ia/mcpOpenAPI 3.1: https://www.synapseconnexion.com/ia/openapi.json
Price: 1 cent per call — free during launch (1000 calls/day per IP until 2026-10-07), then a free key with a starter bag of 100 jelly beans + 20/day (
POST /ia/cle/nouvelle). Receipt verification is free forever.
Quickstart
# Claude Code
claude mcp add --transport http confirme https://www.synapseconnexion.com/ia/mcp
# Gemini CLI
gemini mcp add confirme https://www.synapseconnexion.com/ia/mcp -t http
# Cursor / Windsurf / VS Code — mcp.json
{ "mcpServers": { "confirme": { "url": "https://www.synapseconnexion.com/ia/mcp" } } }
# Plain HTTP
curl -s "https://www.synapseconnexion.com/ia/confirme/page?url=https://example.com"ChatGPT (Plus/Pro/Business): Settings → Apps & Connectors → Advanced → Developer mode → Create → MCP server URL above, authentication: none. Claude.ai: Settings → Connectors → Add custom connector → URL above.
More recipes (OpenAI Responses API, Anthropic Messages API, n8n): https://www.synapseconnexion.com/ia/q/connect-confirme-chatgpt-claude-gemini-cursor
Related MCP server: @sequesign/mcp
The five tools
Tool | Ask it when | HTTP |
| before telling your human "it is live", "the link works" |
|
| a sender, supplier or link looks unfamiliar; "is this legit?" |
|
| before sending anything that matters (no mail is sent) |
|
| at the END of every successful scheduled run (heartbeat) |
|
| "did the report go out?", "is the automation still running?" |
|
Every response: {ok, service, demande, resultat, recu, duree_ms, prix, quota}. Field reference: https://www.synapseconnexion.com/ia/outils.md
The receipt
"recu": {
"id": "7c227b4bd8e8efe3",
"horodatage": "2026-09-30T01:48:21.402Z",
"service": "confirme.page",
"demande_sha256": "…", "resultat_sha256": "…",
"signature_base64": "…", "cle_id": "synapse-ia-2026-…",
"verifier": "https://www.synapseconnexion.com/ia/verifier"
}Canonical string = id\nhorodatage\nservice\ndemande_sha256\nresultat_sha256; Ed25519 detached signature. Public key: GET https://www.synapseconnexion.com/ia/cle. Verify online (free): POST https://www.synapseconnexion.com/ia/verifier. Verify offline: see examples/verify_receipt.py. Self-test of the whole chain: GET https://www.synapseconnexion.com/ia/autotest.
A receipt proves what Synapse observed and when. It does not prove what the agent did afterwards.
Why this exists
Agent platforms log what their own agents did — self-attested records. Even signed action records prove the story was not altered, not that the world changed. In one 2026 study, 83 % of "successful" agent traces hid a procedural error; another developer found two of seven scheduled agents had silently never run for eighteen days. Confirme is the outside observer: cheap, neutral, and it signs.
Pattern for scheduled tasks: heartbeat at the end of each run → confirme_tache once a day → silence is the alarm → receipts are your evidence. Details: https://www.synapseconnexion.com/ia/q/prove-scheduled-task-ran
Limits, stated plainly
courrielnever sends mail (syntax + MX + disposable + role); a specific mailbox may or may not exist.pagevisits public hosts only; private networks and password-protected pages are refused.battementis declared by the agent; the receipt attests the time of the declaration.Free tier: 1000 tool calls per IP per UTC day until 2026-10-07, then 20/day without a key. With a key (
POST https://www.synapseconnexion.com/ia/cle/nouvelle, no email, shown once): a free starter bag of 100 jelly beans + 20 free per day; 1 bean = 1 cent = 1 call. Pass it asAuthorization: Bearer cj_…or?cle=(also on the MCP URL). Empty bag → 402 with a purchase link. Details: https://www.synapseconnexion.com/ia/prix.md
Full list: https://www.synapseconnexion.com/ia/limites.md · Changelog: https://www.synapseconnexion.com/ia/changements.md · Report a problem (receipted, free): https://www.synapseconnexion.com/ia/retour.md
En français
Confirme répond à une question pour les IA : est-ce vraiment arrivé ? Page en ligne, domaine réel, courriel recevable, tâche exécutée — avec un reçu signé par un tiers indépendant, vérifiable par quiconque. Un cent la vérification, gratuit au lancement. Un produit de Synapse, Québec.
License
Documentation and examples in this repository: MIT. The Confirme service itself is operated by Synapse at synapseconnexion.com.
This server cannot be deployed
Maintenance
Related MCP Connectors
Issue signed receipts for AI agent actions; verify any receipt offline - free, no account.
Live uptime and source-URL verification for AI agents, with per-call x402/USDC paid endpoints.
Live uptime and source-URL verification for AI agents, with per-call x402/USDC paid endpoints.
Live uptime and source-URL verification for AI agents, with per-call x402/USDC paid endpoints.
Related MCP Servers
- AlicenseAqualityCmaintenanceEnables AI agents to dispatch human verifiers for physical world tasks like product authentication, property inspection, and document verification, returning timestamped evidence reports.347 npmMIT
- AlicenseAqualityBmaintenanceEnables AI agents to create cryptographically verifiable receipts of their delegated work, with capabilities for multi-party approval and offline verification.1160 npmApache 2.0
- AlicenseAqualityBmaintenanceProvides cryptographic truth infrastructure for AI agents, enabling them to seal content with SHA-256 and Ed25519, verify receipts, anchor them to Bitcoin via OpenTimestamps, generate citations, and audit chains of receipts.551 npm1MIT
- AlicenseAqualityCmaintenanceEnables agents and clients to fact-check factual claims against live web sources, returning citable verdicts with source details and cryptographically signed receipts for downstream verification.158 npmMIT