Agentic RPC
Agentic RPC is the part of Ankr built for software that reads chains without a human in the loop. A coding assistant, an autonomous agent, or a script driven by a model asks a question, and gets an answer already decoded, already sized for a context window, and already impossible to turn into a transaction.
It is three pieces, and they are separate on purpose.
Which one do you need
| You want to | Use | Auth |
|---|---|---|
| Let Claude Code, Cursor, or VS Code read chain data | Agent RPC MCP | your Ankr API key |
| Get smaller, decoded responses from your own HTTP client | TORPC | your Ankr API key |
| Let an agent manage your keys, usage, balance, and alerts | Management MCP | browser sign-in (OAuth) |
| Sign or broadcast a transaction | none of these | a wallet you control |
Nothing on this page can move funds. The data plane refuses broadcast and signing on every chain family, and the control plane manages the account, not a wallet.
The three pieces
Agent RPC MCP is a hosted Model Context Protocol server at https://mcp.ankr.com/rpc. It gives an agent seventeen read-only tools over every chain Ankr serves: transactions and logs with their ABI decoded, wallet and token data from the indexer, contract lookups, and an escape hatch for any read method the routed tools do not cover. One install command and your existing API key.
TORPC is the protocol underneath. One request header on an ordinary JSON-RPC call, and the answer comes back decoded: event logs as named arguments, hex numbers as decimals, and the fields nothing reads dropped. It is an open specification under CC0, not an Ankr feature: w3tech/torpc is the normative document and anyone may implement it. You do not need an agent or an MCP client to use it, and you do not need Ankr.
Management MCP is the control plane: API keys and their security settings, usage and spending, balance and payments, notifications. It is the console, exposed to an MCP client, and it signs you in through a browser rather than taking a key. It lives with the rest of the account documentation because that is what it manages.
Why decoding on the server matters
An agent that calls JSON-RPC directly pays twice: once for hex it cannot read, and again for the reasoning it spends turning that hex into meaning.
A transaction receipt on Ethereum, fetched with one extra header, went from 3,182 bytes to 1,177, a 63% reduction, with its three event logs decoded into named arguments. Read through the MCP server, that same receipt reports a real token count of 530 for the text it emitted, so the agent can budget its own context instead of guessing.
That is the whole thesis: decode once, on the way out, where it is cheap.
Start here
- Quickstart: an API key, one install command, and a first real answer, in about five minutes.
- Agent RPC MCP: every tool, the compression contract, and the read-only guarantee.
- TORPC: the header, the tiers, and a measured before and after.
- Troubleshooting: the failures you hit in the first hour.
Frequently asked questions
How do I connect Claude Code to blockchain data?
One command: claude mcp add --transport http ankr-agent-rpc https://mcp.ankr.com/rpc --header "x-ankr-api-key: <YOUR_KEY>", then /mcp to confirm it connected. The quickstart walks through the first question and the answer it returns.
Do I need an API key, or is there a free mode?
You need an Ankr API key, and there is no keyless mode. The key is the one you already use for rpc.ankr.com: it is passed straight through, so your plan's rate limits and charging apply exactly as they do to your own calls. The server adds no quota and no separate agent plan.
Can an AI agent send a transaction through this?
No. Broadcast and signing are refused in the server, on every chain family, with no exceptions: eth_sendRawTransaction, MEV bundle variants, personal_* and eth_sign*, Solana sendTransaction, Bitcoin sendrawtransaction, Sui sui_executeTransactionBlock, XRPL submit, Tron broadcasttransaction, Cosmos broadcast_tx_*, Starknet add*Transaction. Read-only simulation still works. Sign and send with your own wallet.
Which chains can an agent read?
Raw chain reads reach any chain Ankr serves, EVM mainnets and testnets alike, plus non-EVM families such as Solana, Bitcoin, Sui, XRP, TON, NEAR, Aptos and Cosmos chains. The chains list is the published set. Tier-2 compression is verified on the common EVM chains, and the indexer-backed wallet and token tools need Advanced API coverage, which is a smaller set. Call listChains rather than hardcoding any of it.
Does it work with Cursor and VS Code?
Yes, and with any MCP client that speaks Streamable HTTP. Cursor and VS Code both take a one-click or one-command install, and the same server is on npm as @w3tech.io/agent-rpc-mcp for clients that prefer a local stdio process.
Do I have to use MCP to get the smaller responses?
No. TORPC is one request header on ordinary JSON-RPC, usable from curl or any HTTP client. The MCP server asks for it on your behalf and adds paging, display caps and a token count per response.
What is the difference between Agent RPC MCP and Management MCP?
Agent RPC MCP is the data plane: it reads chains, authenticates with an API key, and cannot change anything. Management MCP is the control plane: it manages your keys, usage, balance and alerts, and signs you in through a browser. Different servers, different auth, no overlap.
How it relates to the rest of Ankr
Agentic RPC is a surface over the same infrastructure, not a separate product. The Node API serves the chains, the Advanced API provides the indexed wallet, token and NFT data, and your key's rate limits and charging apply exactly as they do to your own calls. There is no separate agent quota and no separate agent plan.
For responses that carry a cryptographic proof of the node that produced them, see Verifiable RPC, which is a different guarantee on a different track.