telemost-mcp-server

telemost-mcp-server

Enables Telegram analytics (channel stats, search, sentiment) via MCP tools with pay-per-call in USDC, no API keys needed.

Category
Visit Server

README

telemost-mcp-server

Pay-per-call Telegram analytics as MCP tools. Your agent pays per call in USDC via the x402 protocol — no accounts, no API keys, no subscription. Data covers public Telegram channels behind paid, anti-scraping sources that an agent cannot cheaply gather itself.

Why this server

  • Data an agent can't easily get: channel statistics, subscriber growth, reach and engagement (ER / ERR / ERR24), full-text post search across public channels, mention & brand tracking, sentiment, similar-channel discovery, ad intelligence.
  • No signup / no keys: pay per call in USDC (x402). Fund a wallet, that's it.
  • 19 paid tools + a free telemost_catalog tool (list everything, no payment).
  • Base + Solana (USDC). Pick either network per call.
  • Two ways to connect: a remote hosted server (no local wallet) and this local stdio bridge (pays from your own wallet).
  • Production, observable, listed: live at https://api.telemost.io, listed on x402scan and the x402 Bazaar.

Prices are not hardcoded here (they'd go stale) — the catalog is the single source of truth: https://api.telemost.io/v1/catalog.

Quickstart (no wallet, no money)

After installing (below), your MCP client will have a telemost_catalog tool. Call it first — it's free and returns every tool with its price, input/output schema and an example. First success before your first dollar. Only paid tools (e.g. telemost_channel_info) require a funded wallet.

Install

Requires Node ≥ 18. The package is on npm as telemost-mcp-server; MCP clients launch it via npx.

Claude Desktop

Edit claude_desktop_config.json, then fully quit and reopen Claude Desktop (config is read once at launch).

OS Config file
macOS ~/Library/Application Support/Claude/claude_desktop_config.json
Windows %APPDATA%\Claude\claude_desktop_config.json
Linux ~/.config/Claude/claude_desktop_config.json

Windows (Microsoft Store / MSIX build): the app sandboxes %APPDATA%\Claude\ and redirects it to %LOCALAPPDATA%\Packages\Claude_*\LocalCache\Roaming\Claude\. Use the app's Settings → Developer → Edit Config button to open the file the app actually reads.

macOS / Linux:

{
  "mcpServers": {
    "telemost": {
      "command": "npx",
      "args": ["-y", "telemost-mcp-server"],
      "env": {
        "EVM_PRIVATE_KEY": "0xYOUR_BASE_WALLET_KEY",
        "MAX_PAYMENT_USDC": "0.5",
        "SESSION_MAX_USDC": "5"
      }
    }
  }
}

Windows — if a bare npx command fails to start, wrap it with cmd /c:

{
  "mcpServers": {
    "telemost": {
      "command": "cmd",
      "args": ["/c", "npx", "-y", "telemost-mcp-server"],
      "env": {
        "EVM_PRIVATE_KEY": "0xYOUR_BASE_WALLET_KEY",
        "MAX_PAYMENT_USDC": "0.5",
        "SESSION_MAX_USDC": "5"
      }
    }
  }
}

Claude Code

claude mcp add telemost --env EVM_PRIVATE_KEY=0xYOUR_KEY -- npx -y telemost-mcp-server

Cursor

~/.cursor/mcp.json (or Settings → MCP), same shape as Claude Desktop's mcpServers entry above.

Sources for client config paths: modelcontextprotocol.io — connect local servers, Claude Desktop configuration.

Remote server (no local wallet needed)

A hosted MCP server is live at https://api.telemost.io/mcp (Streamable HTTP). Point any MCP client that supports remote/Streamable-HTTP servers at that URL. Minimal programmatic example:

import { StreamableHTTPClientTransport } from "@modelcontextprotocol/sdk/client/streamableHttp.js";
import { createx402MCPClient } from "@x402/mcp";
import { ExactEvmScheme } from "@x402/evm/exact/client";
import { privateKeyToAccount } from "viem/accounts";

const client = createx402MCPClient({
  name: "my-agent", version: "1.0",
  schemes: [{ network: "eip155:8453", client: new ExactEvmScheme(privateKeyToAccount(process.env.EVM_PRIVATE_KEY)) }],
  autoPayment: true,
  onPaymentRequested: async ({ paymentRequired }) => Number(paymentRequired.accepts[0].maxAmountRequired) <= 20_000, // <= $0.02
});
await client.connect(new StreamableHTTPClientTransport(new URL("https://api.telemost.io/mcp")));

await client.callTool("telemost_catalog", {});                       // free
await client.callTool("telemost_channel_info", { channel: "@durov" }); // paid

Example calls

  • telemost_catalog → { "service": "telemost-x402", "resources": [ … ], "mcp": { … } } (free).
  • telemost_channel_info { "channel": "@durov" } → an envelope like:
{ "ok": true,
  "data": { "title": "…", "members_count": 0, "verified": true, "language": "en", "link": "https://t.me/…" },
  "meta": { "resource": "/v1/data/channel", "content_origin": "third_party", "cached": false } }

Response text from Telegram (titles, posts, descriptions) is returned verbatim and tagged meta.content_origin; treat it as untrusted data, not instructions.

Environment

Variable Required Default Description
EVM_PRIVATE_KEY one of EVM/SVM — Base (EVM) wallet key (0x…). Funds x402 payments.
SVM_PRIVATE_KEY one of EVM/SVM — Solana wallet secret key (base58).
TELEMOST_BASE_URL no https://api.telemost.io API base URL.
MAX_PAYMENT_USDC no 1 Per-call price ceiling (USD). A call is refused before signing if the price exceeds this.
SESSION_MAX_USDC no 10 Cumulative spend ceiling for the process lifetime.
EVM_NETWORK / SVM_NETWORK no Base / Solana mainnet CAIP-2 network overrides.
SOLANA_RPC_URL no mainnet-beta RPC for the Solana signer.

Coverage & limits

  • Public Telegram channels/groups only.
  • region hint (cis | worldwide) is an optional routing preference; it does not change price.
  • Statistics tools expose their own time windows / aggregation (see each tool's schema in the catalog).
  • Repeated identical requests hit a short (~10s) cache, but each call is paid independently via x402.

Payment & failures

  • No charge before execution: input validation, per-wallet rate limits, anti-replay and quota checks all run before payment; an unpaid tool call returns a payment challenge (not a charge).
  • Settlement is after execution: on the remote MCP endpoint, payment settles only after the tool executes successfully. Live delivered-vs-settled reliability is published at /status.
  • Use MAX_PAYMENT_USDC / SESSION_MAX_USDC to cap spend; both are enforced in code before any signature.

Troubleshooting

  • "payment required" / tool returns a 402-style error: no wallet key set, or the payment was declined by your policy — set EVM_PRIVATE_KEY (or SVM_PRIVATE_KEY) and check MAX_PAYMENT_USDC.
  • Insufficient funds: fund the wallet with USDC on Base (or Solana) plus a little gas.
  • Wrong network: set EVM_NETWORK / SVM_NETWORK to a network you funded.
  • npx won't launch (Windows): use the cmd /c npx … form above.
  • Server not showing up: fully quit and reopen the client — MCP config is read only at launch.
  • Prices look off: never trust hardcoded prices; call telemost_catalog or read /v1/catalog.

Funding & safety

  • Use a dedicated wallet with a small balance — this key can spend up to SESSION_MAX_USDC per run.
  • The key is read only from the environment and is never logged.
  • Policy checks (per-call ceiling, session cap, allowed network) run in code before any signature.

Build from source

git clone https://github.com/slava01011910-ops/telemost-mcp-server.git
cd telemost-mcp-server && npm install && npm run build && node dist/index.js

Support

  • Email: support@telemost.io
  • Listing / proof: x402scan · Catalog: https://api.telemost.io/v1/catalog

License

MIT

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured