
Morse — Telegraph in Telegram
io.github.Harshyadav442277由 GitHub 帳號 Harshyadav442277 發布v0.1.0更新於 Oct 24, 2026
Ask Telegraph's ranked miners from any MCP client, no wallet. Morse pays; each answer has a receipt.
概覽
託管式 MCP 端點,把問題送給 Telegraph 的排名礦工並回傳附可驗證收據的答案,不需錢包。
- 可以做什麼
- 提問並取得 Telegraph 排名礦工的回答,收據包含礦工、意圖、排名、費用、延遲與結算交易
- 直接指定某個礦工提問,依你的要求繞過路由器
- 取得問題的配方,並核對回傳的 signal hash 與節點紀錄
- 讀取公開帳本以及意圖、排行榜等免費探索資料
- 適用情境
- 當你想讓助理查詢 Telegraph 礦工網路取得答案,又不想設定錢包、測試網 USDC 或 x402 用戶端時使用;也適合需要每個答案都附可查核收據的情境。
- 執行需求
- 遠端 Streamable-HTTP 端點,不需本機安裝。需要從專案金鑰頁面取得免費 API key,並以 Authorization bearer 標頭送出。需能連線該託管服務。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Morse — Telegraph in Telegram,將其新增到工作區。
- 連接 Authorization token 憑證。它們會被私密儲存,絕不會在公開頁面上展示。
- 為需要使用其工具的對話啟用該服務。
Web executable,透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。
交給代理程式安裝
把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式,或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南,先向你展示發布者、端點和登入方式,經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。
閱讀 https://sourceweft.com/mcp/io.github.Harshyadav442277%2Ftelegraph-morse/install.md,並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前,先告訴我它的發布者、端點和登入方式,等我確認。不要把 API 金鑰寫進命令或我會提交的檔案;問我金鑰應從哪裡取得。其他 MCP 客戶端
把它新增到你客戶端的 mcpServers 設定中。
{
"mcpServers": {
"morse-telegraph-in-telegram": {
"type": "http",
"url": "https://telegraph-morse.vercel.app/mcp",
"headers": {
"Authorization": "Bearer <YOUR_TOKEN>"
}
}
}
}README
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.
Quick start
Claude Code (MCP, no wallet)
Get a free key at /keys, then:
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
curl
Ask one named miner directly (miner authors: your own slug goes here):
Free, no key needed — the discovery endpoints and the ledger:
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
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:
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
keccak256over the payload and reportsverified: 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).
/proofdepends 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
來源:README.md,提交 1908df8
工具
0版本歷史
1- v0.1.0最新Sep 16, 2026
評分
已滿足適用檢查項的 85%
該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。
- 近 30 天內可連通必需25 分
- 已宣告認證方式必需10 分
- 已列出工具15 分
請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。
- 描述必需15 分
- 圖示10 分
- 文件15 分
- 授權條款10 分