Pumpfun Claims Bot MCP Server

PumpFun Claims Bot

---

Looking for interactive monitoring? The telegram-bot supports watch management, group chats, REST API, SSE streaming, and webhooks. Use this channel-bot for simple broadcast-only channels.

Features

Feed Types

Claim Intelligence

Every GitHub social fee claim card includes:

Graduation Cards

Rich graduation cards include creator profile, top holders analysis, 24h trading volume, dev wallet activity, pool liquidity, and bundle detection.

Influencer Tier System

Combines GitHub and X/Twitter follower data to classify claimers:

Either threshold triggers the tier — a user with 50K X followers and 50 GitHub followers still qualifies as Influencer.

Credibility Score & Dev Track Record

Every claim card leads with a deterministic 0-100 credibility verdict synthesised from the trust signals the bot already gathers, plus the claiming dev's persistent track record across prior tokens.

The score is deterministic and transparent — same inputs always produce the same number, no model call, and every point is attributed to a named factor shown under the headline:

Scoring starts at a neutral 50 and adjusts by weighted factors: claim verification (+22, or -20 on a GitHub-owner mismatch), account age (+14 for 5y+ down to -18 for < 30 days), public repos, followers, starred non-fork repo, copycats (-10/-16), bundling, holder concentration, prior rugs, and bans.

Dev track record turns that one-shot score into memory. Every score is recorded against the claiming GitHub user id and persisted, so the next time that dev launches a token the card shows their history — exposing a serial fee-farmer whose newest coin looks clean, and crediting a builder with a real record:

🟢 Credibility: 100/100 · Strong
   ↑ claim verified · GitHub 6y · 42 repos · repo 150★
📊 Dev track record: 3 prior tokens · avg 🔴 19/100 (High Risk)

Logic lives in src/credibility.ts (pure scoreCredibility) and src/dev-reputation.ts (persistent store), both fully covered by tests.

AI-Powered Claim Summaries

First-time claims get a one-line AI take generated by Groq (llama-3.3-70b-versatile):

• Analyzes 30+ signals: token metadata, GitHub profile, creator history, trading activity
• Max 120 characters — direct, opinionated, trader-focused
• 5-second timeout with graceful fallback (empty string on failure)
• HTML-sanitized output safe for Telegram

Examples:

"Real GitHub project, dev claimed fast" · "Fork of popular repo, proceed with caution"

Architecture

Solana RPC (WebSocket + HTTP polling)
        │
        ▼
┌───────────────────┐
│  SocialFeeIndex   │──▶ Bootstraps ~148K SharingConfig → mint mappings
└────────┬──────────┘
         │
┌────────▼──────────┐
│   ClaimMonitor    │──▶ Decodes PumpFees program claim transactions
│   EventMonitor    │──▶ Decodes Pump program logs (graduations)
└────────┬──────────┘
         │ FeeClaimEvent / GraduationEvent
┌────────▼──────────┐
│ Enrichment Layer  │
│  ├─ GitHub API    │──▶ User profile, repos, followers
│  ├─ X/Twitter API │──▶ Follower counts, influencer tier
│  ├─ PumpFun API   │──▶ Token info, creator profile, holders, trades
│  ├─ ClaimTracker  │──▶ First-claim detection, persistent counts
│  └─ Fake Detect   │──▶ Instruction called but no payout (amountLamports=0)
└────────┬──────────┘
         │ ClaimFeedContext
┌────────▼──────────┐
│    Formatters     │──▶ Rich HTML cards with sections & emoji layout
└────────┬──────────┘
         │
┌────────▼──────────┐
│   grammY Bot      │──▶ Posts photo + caption to Telegram channel
│   (retry + rate   │    Falls back to text-only if photo fails
│    limiting)      │
└───────────────────┘

Multi-RPC Failover

The RpcFallback class manages multiple Solana RPC endpoints with automatic rotation:

• Round-robin rotation after 3 consecutive failures on any endpoint
• 60-second cooldown per failed endpoint before retrying
• Earliest-expiry fallback — if all endpoints are in cooldown, picks the one expiring soonest
• Success resets — a single successful call resets the failure counter
• Configure via SOLANA_RPC_URLS (comma-separated)

Programs Monitored

Install

npm (MCP server — recommended for AI assistants)

Run the MCP server instantly with npx — no clone needed:

npx pumpfun-claims-bot

Or install globally:

npm install -g pumpfun-claims-bot
pumpfun-claims-bot

Add to your MCP client config (Claude Desktop, Cursor, VS Code Copilot):

{
  "mcpServers": {
    "pumpfun": {
      "command": "npx",
      "args": ["pumpfun-claims-bot"]
    }
  }
}

npm package: pumpfun-claims-bot on npm

From source (full bot + Telegram feed)

git clone https://github.com/nirholas/pumpfun-claims-bot.git
cd pumpfun-claims-bot && npm install

Quick Start

1. Create a Telegram Bot

1. Message @BotFather on Telegram
2. /newbot → follow prompts → copy the bot token
3. Create a public channel (e.g., @pumpfunclaims)
4. Add the bot as an admin to the channel (must have "Post Messages" permission)

2. Configure Environment

cp .env.example .env

# ── Required ──────────────────────────────────────────────
TELEGRAM_BOT_TOKEN=your-bot-token-from-botfather
CHANNEL_ID=@your_channel_name    # or numeric chat ID like -100xxx

# ── Solana RPC ────────────────────────────────────────────
SOLANA_RPC_URL=https://mainnet.helius-rpc.com/?api-key=your-key
SOLANA_WS_URL=wss://mainnet.helius-rpc.com/?api-key=your-key

# Comma-separated fallback RPCs (rotates on 429/5xx/timeout)
SOLANA_RPC_URLS=https://mainnet.helius-rpc.com/?api-key=key1,https://your-other-rpc.com

# ── Feed Toggles (all default false except FEED_CLAIMS) ───
FEED_CLAIMS=true                 # GitHub social fee claims (default: true)
FEED_GRADUATIONS=false           # Token graduations to PumpAMM (default: false)
FEED_LAUNCHES=false              # New token launches (default: false)
FEED_WHALES=false                # Large buy/sell whale alerts (default: false)
FEED_FEE_DISTRIBUTIONS=false     # Creator fee distribution events (default: false)

# ── Claim Filter ──────────────────────────────────────────
REQUIRE_GITHUB=true              # Only post claims with GitHub social fee PDA (default: true)

# ── Enrichment APIs ───────────────────────────────────────
GITHUB_TOKEN=ghp_your_token      # Raises GitHub rate limit: 60 → 5000 req/hr
GROQ_API_KEY=gsk_your_key        # Groq API for AI one-liner summaries

# X/Twitter follower counts & influencer detection
# Get cookies from x.com DevTools → Application → Cookies
# X_AUTH_TOKEN=your_auth_token_cookie
# X_CT0_TOKEN=your_ct0_cookie

# ── Affiliate Ref Codes ───────────────────────────────────
# Appended to Axiom / GMGN / Padre trading links in cards
# AXIOM_REF=your_ref
# GMGN_REF=your_ref
# PADRE_REF=your_ref

# ── Tuning ────────────────────────────────────────────────
POLL_INTERVAL_SECONDS=30         # HTTP polling fallback interval (default: 30)
WHALE_THRESHOLD_SOL=10           # Minimum SOL for whale alerts (default: 10)
LOG_LEVEL=info                   # debug | info | warn | error (default: info)

# ── Health Check ──────────────────────────────────────────
# PORT=3000                      # Set automatically by Railway

Environment Variables Reference

3. Run

# Install dependencies
npm install

# Development (hot reload via tsx)
npm run dev

# Production
npm run build
npm start

4. Deploy with Docker

docker build -t pumpfun-channel-bot .
docker run -d --env-file .env pumpfun-channel-bot

5. Deploy to Railway

Railway auto-deploys from GitHub and provides persistent volumes for claim tracking data.

# Install Railway CLI
npm install -g @railway/cli
railway login

# Create & link project
railway init
railway link

# Set environment variables
railway variables set TELEGRAM_BOT_TOKEN=your-token
railway variables set CHANNEL_ID=@your_channel_name
railway variables set SOLANA_RPC_URL=https://mainnet.helius-rpc.com/?api-key=your-key
railway variables set SOLANA_WS_URL=wss://mainnet.helius-rpc.com/?api-key=your-key
railway variables set FEED_CLAIMS=true
railway variables set REQUIRE_GITHUB=true

# Create persistent volume for claim tracker data
railway volume create --mount /app/data

# Deploy
railway up

See railway.json for the deployment config:

{
  "$schema": "https://railway.app/railway.schema.json",
  "build": {
    "builder": "DOCKERFILE",
    "dockerfilePath": "Dockerfile"
  },
  "deploy": {
    "restartPolicyType": "ON_FAILURE",
    "restartPolicyMaxRetries": 10
  }
}

Project Structure

pumpfun-claims-bot/
├── src/
│   ├── index.ts              # Entry point — wires monitors, enrichment, & Telegram posting
│   ├── config.ts             # Environment variable loading & validation
│   ├── mcp-server.ts         # MCP server — exposes tools via Streamable HTTP or stdio
│   ├── mcp-stdio.ts          # Standalone MCP entry point (stdio transport)
│   ├── claim-monitor.ts      # PumpFees program monitor (WebSocket + HTTP polling)
│   ├── claim-routing.ts      # Decides which claims are posted, skipped, or re-routed
│   ├── claim-tracker.ts      # First-claim detection + claim counter (persisted to disk)
│   ├── credibility.ts        # Deterministic 0-100 credibility scoring
│   ├── dev-reputation.ts     # Persistent per-developer track record store
│   ├── event-monitor.ts      # Pump program log decoder (graduations, launches)
│   ├── social-fee-index.ts   # SocialFeeIndex — maps SharingConfig PDAs → mints (~148K)
│   ├── formatters.ts         # Rich HTML card builders for Telegram
│   ├── pump-client.ts        # PumpFun HTTP API client (token info, creator profiles)
│   ├── github-client.ts      # GitHub API client (user profiles, rate-limited cache)
│   ├── x-client.ts           # X/Twitter profile fetcher + influencer tier logic
│   ├── groq-client.ts        # Groq AI one-liner summaries
│   ├── rpc-fallback.ts       # Multi-RPC failover with round-robin
│   ├── health.ts             # HTTP health check server
│   ├── types.ts              # Program IDs, discriminators, event types
│   └── logger.ts             # Leveled console logger
├── src/__tests__/            # Vitest suites (203 tests) + shared fixtures
├── packages/web/             # React dashboard, mock-data build (see Web Dashboard)
├── web/web/                  # React dashboard, live-SSE build (see Web Dashboard)
├── leaderboard-bot/          # Separate Telegram bot: GitHub dev earnings leaderboard
├── data/                     # Persisted state (gitignored, Railway volume mount)
│   └── github-first-claims.json
├── Dockerfile                # Multi-stage Docker build
├── railway.json              # Railway deployment config
├── package.json
└── tsconfig.json

How It Works

Claim Detection Pipeline

Transaction detected on PumpFees program
  │
  ▼
Identify instruction: claim_social_fee_pda?
  │
  ├─ YES ──▶ Parse platform (2 = GitHub) + user_id from Anchor args
  │           │
  │           ▼
  │        Check amountLamports from SocialFeePdaClaimed event
  │           │
  │           ├─ amountLamports > 0 ──▶ Real claim
  │           │   ├─ Check ClaimTracker: first time for this GitHub user?
  │           │   │   ├─ YES ──▶ 🚨 FIRST TIME CLAIM banner
  │           │   │   └─ NO  ──▶ Standard claim card
  │           │   └─ Enrich: GitHub API + PumpFun API + X profile
  │           │
  │           └─ amountLamports = 0 ──▶ ⚠️ FAKE CLAIM (instruction called, no payout)
  │
  └─ NO ───▶ Other claim type (creator fee, cashback, etc.)

SocialFeeIndex Bootstrap

On startup, the bot fetches all SharingConfig accounts from the PumpFees program to build a reverse mapping from social fee PDA addresses to token mints. This enables resolving which token a social fee claim belongs to without additional RPC calls.

• ~148K mappings loaded at startup
• Incremental updates via WebSocket subscription on CreateFeeSharingConfig and UpdateFeeShares events
• Lookup: socialFeeIndex.getMintForPda(pdaAddress) → token mint
• One PDA → many mints — handles scammers who reuse PDAs across multiple tokens

Fake Claim Detection

Some users call the claim_social_fee_pda instruction targeting random token PDAs where they have no fees to collect. The bot detects these by checking:

1. The instruction discriminator matches claim_social_fee_pda
2. The transaction logs contain no SocialFeePdaClaimed event — OR the event shows amountLamports = 0
3. The GitHub user ID and platform are still parsed from the instruction args (Anchor Borsh format)

Fake claims are posted with a ⚠️ FAKE CLAIM warning and a 🚩 Fake claim — no fees paid out trust signal.

First-Claim Tracking

The ClaimTracker maintains a persistent set of GitHub user IDs that have successfully claimed:

• In-memory set for fast lookup during processing
• Debounced disk persistence (5-second delay) to data/github-first-claims.json
• Split check/mark pattern: hasGithubUserClaimed() checks without side effects, markGithubUserClaimed() only called after successful Telegram post
• Claim counter: incrementGithubClaimCount() returns sequential claim number per user
• First-claim status is NOT set for fake claims

Three-Layer First-Claim Verification

1. Local dedup — Skip if already posted (survives restarts via persisted JSON)
2. On-chain verification — lifetimeClaimedLamports == amountLamports confirms it's truly the first claim on-chain
3. Graceful fallback — Skips first-claim banner if verification fails (prevents false positives after redeployment)

Example Claim Card

🚨🚨🚨 FIRST TIME CLAIM 🚨🚨🚨

🐙 $PUMP — PumpCoin  💹 $45K
  ↳ GitHub dev claimed PumpFun social fees

📊 Claim #1 · 0.1043 SOL lifetime ($15.65)

🏦 0.1043 SOL ($15.65)
  ↳ 8mNp...4rWz

👤 nirholas (Nicholas)
  ↳ 📦 45 · 👁 200 · 📅 5y ago
  TypeScript SDK builder
𝕏 nichxbt · 1.2K

📈 Bonding curve (72%) · Created 3h ago · 💬 12
𝕏 @pump_coin · 💬 TG · 🌐 pumpcoin.io

⚠️ GitHub account created 15d ago

CA: 7xKXt...p3Bz
Axiom · GMGN · Padre

🔍 TX

Requirements

• Node.js >= 20.0.0
• Telegram bot token (via @BotFather)
• Telegram channel with the bot added as admin
• Solana RPC endpoint — dedicated RPC recommended (Helius, QuickNode, Triton). Public mainnet works but may rate-limit.
• GitHub token (optional) — raises API rate limit from 60 to 5,000 req/hr

Troubleshooting

Bot Not Posting Messages

1. Check bot permissions — The bot must be an admin in the channel with "Post Messages" permission
2. Verify CHANNEL_ID — Use @channel_name for public channels or the numeric ID (e.g., -100xxx) for private channels. To find the numeric ID, forward a channel message to @userinfobot
3. Telegram 403 error — Means the bot is NOT a member/admin of the channel. Add it via channel settings → Administrators → Add Administrator
4. Check logs — Set LOG_LEVEL=debug to see all events the bot processes

Rate Limiting

Telegram limits bots to ~30 messages per second to a channel. The grammY framework handles rate limiting automatically:

• Messages may be delayed but won't be dropped
• The bot includes a retry helper that respects retry_after headers
• For very high activity, increase POLL_INTERVAL_SECONDS to reduce event volume

RPC Connection Issues

• Public RPC endpoints have rate limits — for production use a dedicated RPC
• Set SOLANA_RPC_URLS with multiple endpoints for automatic failover
• If WebSocket disconnects, the bot falls back to HTTP polling at POLL_INTERVAL_SECONDS
• The RpcFallback class provides round-robin across configured endpoints
• Set LOG_LEVEL=debug to see connection status

Missing Claims

• GitHub claims only? — Set REQUIRE_GITHUB=true to only post GitHub social fee claims
• Feed disabled? — Verify FEED_CLAIMS=true is set
• SocialFeeIndex slow? — Initial bootstrap fetches ~148K accounts. This takes 30-60 seconds on startup. Check logs for SocialFeeIndex: loaded N mappings
• Rate limited? — GitHub API allows 60 req/hr unauthenticated. Set GITHUB_TOKEN for 5,000 req/hr

Pipeline Stats

The bot logs pipeline counters every 60 seconds:

Pipeline: 15 total → 8 social → 3 first / 5 repeat → 8 posted (skip: 7 cashback)

• total: All claim events received
• social: GitHub social fee PDA claims
• first/repeat: First-time vs. returning claimers
• posted: Successfully posted to Telegram
• skip cashback: Cashback claims (user refunds, not creator activity)

Web Dashboard

Two React frontends are checked in, and they are not the same app:

The features listed below describe web/web/. Both directories build clean today; consolidating them onto one is still open work.

Deep links (/dashboard, /docs, ...) are client-side routes, so any static host must rewrite unknown paths to index.html. Each app now ships a vercel.json with that rewrite; on a non-Vercel host, configure the equivalent SPA fallback.

Pages

Dashboard Features

• Server-Sent Events (SSE) — Real-time streaming via /api/v1/claims/stream with auto-reconnect (3s delay)
• Event Filters — All, Launches, Whales, Graduations, Claims, Distributions
• Live Stats Bar — Counters for each event type
• Watch Lists — Add/remove wallet addresses to monitor (GET/POST/DELETE /api/v1/watches)
• Rich Event Cards — Token launches, whale trades, graduations, fee claims with full context
• Connection Status — Visual indicator with error messages
• Mock Data Fallback — Simulated feed when SSE is unavailable

Tech Stack

MCP Server

The bot includes a built-in Model Context Protocol (MCP) server that lets AI assistants (Claude, Copilot, Cursor, etc.) query PumpFun on-chain data conversationally.

MCP Tools

Usage: Stdio Transport (Claude Desktop / Cursor / VS Code)

Add to your MCP client config (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "pumpfun": {
      "command": "npx",
      "args": ["pumpfun-claims-bot"]
    }
  }
}

Or run from source:

# Development (tsx)
npm run mcp:dev

# Production (compiled)
npm run build && npm run mcp

Usage: Streamable HTTP Transport (Embedded)

Run alongside the main bot by setting MCP_ENABLED=true:

MCP_ENABLED=true MCP_PORT=3001 npm run dev

The MCP endpoint is available at POST /mcp on the configured port. Clients connect using the Streamable HTTP transport.

Example Queries

Once connected, ask your AI assistant:

• "Look up token info for mint address 7xKXt...p3Bz"
• "Has GitHub user 12345 ever claimed PumpFun fees?"
• "Who are the top holders of this token?"
• "Check if this token launch was bundled"
• "What's the current SOL price?"

Health Check API

The bot exposes an HTTP health check server for Railway / Docker probes.

Response:

{
  "status": "ok",
  "uptime": "12345s",
  "uptimeMs": 12345000
}

• Returns 200 for ok, 503 for degraded
• Dynamic stats injected via callback (pipeline counters, connection status)
• Port configurable via PORT or HEALTH_PORT env vars (default: 3000)

Testing

The project uses Vitest: 203 tests across 11 suites, all green on npm test.

# Run all tests
npm test

# Watch mode (re-runs on file changes)
npm run test:watch

Test Suites

Every suite lives in src/__tests__/ (plus fixtures.ts, shared sample data).

Local Development

# Install dependencies
npm install

# Run with hot reload (tsx)
npm run dev

# Type-check without emitting
npm run typecheck

Set LOG_LEVEL=debug — all events are logged to stdout regardless of whether they're posted to Telegram.

Tech Stack

Contributing

Contributions are welcome! See CONTRIBUTING.md for guidelines.

1. Fork the repo
2. Create a feature branch (git checkout -b feat/my-feature)
3. Run tests (npm test) and typecheck (npm run typecheck)
4. Open a Pull Request

Security

Found a vulnerability? Please report it responsibly — see SECURITY.md.

License

All rights reserved. See LICENSE.

Documentation

Full documentation site: https://nirholas.github.io/pumpfun-claims-bot/

• Getting started covers install and first run.
• Examples has copy-paste snippets.