Morse — Telegraph in Telegram

A Telegraph Hackathon Season I, Track 3 application. Live: https://telegraph-morse.vercel.app · public ledger · API & MCP

Telegraph in Telegram. Ask, get an answer from a ranked miner, with a receipt. Send a question to the bot — no wallet, no key, no sign-up — and Telegraph's own router classifies it and picks a ranked miner (Morse falls back to its own routing only when the router does not answer), Morse pays the x402 fee from one app-owned wallet, and the answer comes back with a receipt: the miner that served it, the intent and why it was chosen, that miner's rank, its confidence, the cost, the latency, the on-chain settlement transaction, and a signal_hash you can verify on the node. The same thing is on the web and behind one hosted MCP/REST endpoint for agents.

Live example, 2026-09-02: LiveCert #1 for SSL_VERIFICATION, $0.01, signal 0x0691ca3f…0821a1, settled on-chain as 0x31b9b480…2af007.

Every call Morse makes is in a public ledger, so "people used it" is checkable rather than claimed.

What Morse deliberately does not do. It does not re-rank miners. Telegraph's leaderboard is the consensus and the router's pick is the answer, so there is no podium, no automatic second opinion and no consensus report. Those were built, measured and retired on 2026-09-04 after an organizer pointed out that paying several miners per question to re-derive the ranking is what the protocol already does once for everyone, does not work economically for a user, and looks like spam. The ledger keeps their rows, labelled; the code leaves before the freeze (GAPS G32).

One check that does not rely on Morse's word. /proof reads the payer wallet's USDC transfers on Base Sepolia from a public indexer and matches them, hash for hash, against the ledger's settlement transactions — the "actual volume of Telegraph calls" counted from the chain, with any settlement the ledger lacks shown rather than hidden. It is computed from data that already exists and spends nothing.

Not an aggregator. Morse does not pick providers or blend answers. Telegraph's ranked router picks the miner; Morse pays and keeps the receipt. When you do want a specific miner — the dispatch the organizers' own reference apps use — name it: telegraph_ask_miner over MCP, {"miner": "livecert", "question": "…"} on REST, /miner livecert <question> in Telegram. The receipt says routing was bypassed at your request.

Why

Consuming Telegraph today needs a wallet, testnet USDC from a faucet, and an x402 client. That keeps out everyone who is not already a developer with a burner key, and it means the network's "real usage" is mostly machines calling one miner. Morse is the missing front door, and because every answer carries a verifiable receipt, its usage is evidence rather than a claim.

Related MCP server: Tenjin MCP

Quick start

Claude Code (MCP, no wallet)

Get a free key at /keys, then:

claude mcp add --transport http morse https://telegraph-morse.vercel.app/mcp --header "Authorization: Bearer morse_YOURKEY"

Then ask Claude anything the network can answer — "use telegraph_ask: is the TLS certificate for github.com valid, and who issued it?", then "use telegraph_verify_signal on that signal_hash". Tools: telegraph_ask, telegraph_ask_miner, telegraph_recipe, telegraph_verify_signal. Intents and leaderboards are free over REST at /v1/intents and /v1/leaderboard/{INTENT}; the three MCP discovery tools that duplicated Telegraph's explorer were removed on 2026-09-05.

Cursor, or any Streamable-HTTP MCP client

{
  "mcpServers": {
    "morse": {
      "url": "https://telegraph-morse.vercel.app/mcp",
      "headers": { "Authorization": "Bearer morse_YOURKEY" }
    }
  }
}

curl

curl -X POST https://telegraph-morse.vercel.app/api/keys -H "content-type: application/json" -d '{"label":"my-laptop"}'

curl -X POST https://telegraph-morse.vercel.app/v1/ask -H "Authorization: Bearer morse_YOURKEY" -H "content-type: application/json" -d '{"question":"What is the current weather in Chennai?"}'

Ask one named miner directly (miner authors: your own slug goes here):

curl -X POST https://telegraph-morse.vercel.app/v1/ask -H "Authorization: Bearer morse_YOURKEY" -H "content-type: application/json" -d '{"miner":"livecert","question":"Is the SSL certificate for github.com valid?"}'

Free, no key needed — the discovery endpoints and the ledger:

curl -s https://telegraph-morse.vercel.app/v1/intents

curl -s https://telegraph-morse.vercel.app/v1/leaderboard/SSL_VERIFICATION

curl -s https://telegraph-morse.vercel.app/api/stats

Telegram

https://t.me/MyMorse_Bot — /start shows tappable example questions; answers free text. Commands: /miner <slug> <question>, /safe, /wallet, /weather, /fact, /verify, /stats. Every answer carries the same receipt the web and API surfaces return, and lands in the same public ledger. In a group, Morse answers only when it is @mentioned or replied to, so a busy chat never pays for answers nobody asked for.

What a receipt contains

/verify/{hash} shows the node's record, the payer wallet and whether it is Morse's, the node's own keccak256-over-payload attestation, and the payload the hash covers. The node does not publish a per-call settlement transaction (0 of 8 user-paid signals sampled on 2026-09-02 carried one), so the on-chain trail is the payer wallet's USDC history on BaseScan, which Morse links.

Run it yourself

git clone https://github.com/Harshyadav442277/telegraph-morse && cd telegraph-morse && npm ci

npm run typecheck && npm test

cp .env.example .env && npm run dev

Without EVM_PRIVATE_KEY and a non-zero DAILY_BUDGET_CALLS, asking is refused with an honest message; the site, the ledger and the free discovery endpoints all still work. The end-to-end judge journey runs against any deployment:

npm run e2e

That suite is free to run. The single paid step is gated behind MORSE_E2E_PAID=1 so no schedule can manufacture traffic.

Docs

• GO-LIVE.md — the operator's runbook for the wallet, the bot and the first paid call

• PLAN.md — claim, reality checks, judging criteria, schedule, tooling

• PRD.md · ARCHITECTURE.md · PHASES.md

• GAPS.md — what is missing, broken, or unverified (read before trusting a claim)

• MEMORY.md — decisions and lessons

• DEMO.md — the judge journey with exact expected output

• docs/TELEGRAPH_FACTS.md — protocol facts, sourced and dated

Assumptions and limitations

• Testnet. Base Sepolia, testnet USDC. Answers are real; the money is not.

• Telegraph routes first; Morse only falls back. The network's router gets 20 seconds; if it does not answer (it timed out at ~47s for a day on 2026-09-02), Morse classifies the intent with keyword rules and calls the best-ranked miner directly. Every receipt and ledger row says which of the two happened (GAPS G17).

• The ledger's early rows are our own verification calls, not users — real and receipted, but not adoption (GAPS G20).

• "Users" means distinct salted identity hashes — a Telegram user id, a web session cookie, or an API key. One person on two surfaces counts twice; ten people reading one forwarded answer count once. Morse publishes the method next to the number and never rounds it up (GAPS G4).

• The signal hash is shown, not re-derived. The node states it is keccak256 over the payload and reports verified: true; eleven serialisations of the payload as served failed to reproduce it, so Morse displays that attestation rather than claiming to have recomputed it (GAPS G3). What Morse establishes independently is the payer: the wallet on the record, checked against its own.

• Confidence is heterogeneous. Miners report it in different shapes or not at all; Morse normalises what it can and says "not reported" otherwise (GAPS G8).

• A direct call to a named miner may not be possible for every miner. Direct calls are built from the miner's manifest and declared input schema, and some miners reject a request they did not shape themselves — which fails honestly and costs nothing (GAPS G14, G30).

• /proof depends on a third-party indexer. It reads Blockscout, which can lag the chain by a few blocks or be down; the page then says so and shows the ledger alone. It proves payments, not people. Two settlements on chain have no ledger row — calls Morse recorded as timed out that the network settled anyway — and are listed rather than removed (GAPS G29).

• Nothing is written on chain by Morse. Payments settle on Base Sepolia through x402; Morse reads that trail but does not anchor verdicts or run ERC-8183 jobs (GAPS G27).

License

MIT