Private TON infrastructure
The private lane to TON.
Single-tenant liteservers, full-history archive nodes, a mempool stream and an MCP server for AI agents. Your own ADNL endpoint, your own key, no shared rate limits. Live in under a minute.
What we run for you
Six products. One private backbone.
Everything below runs on our own fleet — pre-synced, monitored every second, answering over native ADNL.
Private liteservers
Low-latency query endpoints with guaranteed throughput, from 10 to 500 req/s. The workhorse behind wallets, bots and mini apps.
Archive nodes
Every block since May 2021. Account state at any height, get-methods against any moment of history.
Mempool stream
Pending external messages the moment our nodes see them — hundreds of milliseconds before the block lands. Built for trading.
Priority broadcast
Your transaction fanned out through nodes in several regions at once, so it lands in the earliest possible block.
Webhooks
Address activity, jetton transfers, NFT sales — pushed straight to your backend. No polling loops to babysit.
MCP for AI agents
Claude, Cursor or any MCP client reads balances, jettons, NFTs and history through one authenticated endpoint.
The full catalog
Public vs private
The global config is free. Here's what it costs you.
TON's own docs recommend a dedicated liteserver for production. Below — a live simulation of the same getAccountState query against both.
→ Failover libraries exist because public liteservers fail. Skip the failure.
Built for real work
The workloads this backbone was built for
High-speed DEX trading
Arbitrage and sniping on STON.fi or DeDust live and die by the round-trip. A dedicated ADNL lane, the mempool stream and priority broadcast buy you the blocks everyone else misses.
Telegram Mini Apps & bots
A viral mini app is a rate-limit incident waiting to happen. Guaranteed throughput keeps a million-user day from ending in error 228.
Wallets
Balances and history have to render before the user blinks — from Berlin and from Singapore alike. Regional endpoints with client-side failover keep p99 flat.
Indexers & explorers
Backfilling three billion transactions through public nodes is a month of retries. Archive depth plus pay-per-request pricing was made for backfills.
AI agents
Agents ask in bursts — 40 calls to trace one jetton transfer, then silence. The MCP server absorbs the spikes and answers from archive depth when asked about 2023.
Compliance & research
Prove what an address held on any date, reconstruct a token flow, audit a treasury. Historical state at any block, exportable.
Pricing
Any load. Any duration. One flat logic.
From a one-cent request to a 400 rps archive endpoint, from a week to a year — and anything shorter is pay-per-request. Assemble exactly what your workload needs, pay in GRAM or USDT.
Minimum term — one week. Need a node for an evening or a load test? That's what pay-per-request is for: no timers, no renewals. Cancel a subscription any time — unused time returns to your balance.
- node
- Lite
- throughput
- 60 req/s
- period
- Month
- Private ADNL access — no shared queues
- Mainnet + testnet configs
- 3 endpoints for client-side failover
- Mempool stream & webhooks add-ons
- Provisioned in under 60 seconds
For spiky and unpredictable loads
- Bursts up to 500 req/s
- Current-state queries
- No monthly commitment
- Perfect for launches and load tests
For backfills and history-heavy jobs
- Bursts up to 500 req/s
- Full-history queries since genesis
- No monthly commitment
- Made for indexer backfills
◆ Top up your dashboard balance in GRAM or USDT — requests draw it down. No monthly fee, and the balance never expires.
A whole machine that answers only to you — the tier serious teams end up on.
16 cores, 64 GB RAM, NVMe Gen4. Your Ed25519 key, your static IP, zero neighbors.
20 TB of chain history behind a single-tenant ADNL endpoint. 99.95% SLA, read replicas on request.
Multi-region clusters, custom hardware, invoices and bank transfer. Tell us the load profile.
◆ All plans payable in GRAM or USDT — pay-in-GRAM gets a 5% discount.
Need a cluster, custom regions or an invoice? Talk to us
For AI agents
Give your agent a node, not a rate limit.
Agents don't query politely. They burst — 40 calls to trace one jetton transfer, then silence, then 40 more. That pattern dies on 1-RPS gateways. The TONNode MCP server plugs Claude and any MCP client straight into our private liteservers and archive nodes.
{
"mcpServers": {
"ton": {
"command": "npx",
"args": ["-y", "@tonnode/mcp"]
}
}
}One block in your MCP config. Your agent is reading mainnet before you finish your coffee.
- 50k requests /mo
- Recent state only
- 1 MCP key
- 2M requests /mo
- Archive queries included
- 5 keys · burst-friendly
- 20M requests /mo
- Dedicated liteserver backend
- 99.9% SLA · unlimited keys
Why private
The parts of TON infrastructure you stop thinking about
Your endpoint. Your key.
A single-tenant liteserver with its own IP, port and Ed25519 key. Drop it into your config and speak native ADNL — no HTTP proxy in the middle.
No 1-RPS ceiling. No error 228.
Public gateways throttle anonymous traffic to 1 request per second; shared plans cap out at 50 RPS on sliding windows. Your server has no meter.
Every block since genesis
Public liteservers drop history past recent blocks. Our archive nodes hold the full chain — May 2021 to the current block — for indexers, explorers and analytics.
Close to your backend
Amsterdam, Frankfurt, Helsinki, Singapore, Ashburn. Pick a region at checkout, or spread endpoints across three for client-side failover.
Watched around the clock
Out-of-sync alarms fire at 10 seconds — well inside the 20-second health bar. If a node drifts, we rotate you to a hot standby before your retries notice.
Sync took us weeks. It takes you a minute.
Restoring an archive node from the official dump means 16 TB of ZFS and 1–2 weeks of sync. We keep pre-synced fleets warm; your endpoint goes live in under 60 seconds.
How it works
From wallet to endpoint in three steps
Connect your wallet
Sign in with TON Connect. No email forms, no KYC for standard plans.
Pick a plan
Shared, dedicated or archive. Pay in GRAM or USDT — one transaction.
Get your config
IP, port and public key — a drop-in liteserver entry for your global config. Paste it into tonutils-go or ton-lite-client and go.
FAQ
Questions engineers actually ask
How does pay-per-request work?
You top up your dashboard balance in GRAM or USDT; every request draws it down — $0.01 on lite, $0.02 on archive, with bursts up to 500 req/s. No monthly fee, and the balance never expires. When it runs low, we warn you in Telegram before anything throttles.
Can I get a node for just a day or a few hours?
Yes — that's exactly what pay-per-request is for. Top up the balance, run your load test or airdrop day, and pay only for the requests you actually made — no timers, no subscriptions to remember to cancel. Subscriptions start at one week; cancel any time and the unused time returns to your balance.
What exactly is the mempool stream?
TON's mempool lives at the node level: external messages sit as pending until validators seal them into a block. Public gateways don't expose that layer at all — our nodes stream every pending message they see over WebSocket, typically a few hundred milliseconds before the block lands. Trading desks use it to react before everyone else.
What exactly is a liteserver?
A TON full node with the liteserver endpoint enabled. It serves raw chain state — accounts, blocks, contract get-methods — over TON's native ADNL protocol. Your client verifies every response with Merkle proofs, so you trust math, not us. Even HTTP APIs like toncenter are proxies sitting on liteservers underneath.
Why not just use the free public config?
For a weekend project — sure. In production you'll meet “failed to send query to server: not ready”, ADNL timeouts, and servers so stale your client can't read their version. No SLA, no archive depth. TON's own docs tell production teams to run a dedicated liteserver. We run it for you.
How fast is provisioning?
Shared and dedicated liteservers: under 60 seconds — the fleet is pre-synced. Archive nodes: usually under an hour; if a warm node isn't free in your region, we tell you the exact ETA before you pay. Compare that to 1–2 weeks restoring the 16 TB official dump yourself.
What payment methods do you accept?
GRAM and USDT (TON network) via TON Connect — GRAM is the renamed Toncoin, plus card and invoice for annual dedicated and archive contracts. Pay-in-GRAM gets a 5% discount.
Do you log my queries?
We keep aggregate counters — request volume, error rates — for billing and capacity planning, retained 30 days. We don't log query payloads, the addresses you look up, or client IPs beyond abuse protection. On dedicated instances, the box serves only you.
What's the SLA? Can I get a refund?
Dedicated and archive plans carry a written 99.9% monthly uptime SLA. Miss it and service credits are applied automatically — no ticket required. Cancel any monthly plan in the first 7 days for a full refund, in GRAM or however you paid.
Stop sharing your liteserver.
Connect a wallet and get a private endpoint today.