Track 3 · Programmable3 of 4
Built for AI agents
XChain treats AI agents as first-class users: read everything without a key, connect over the Model Context Protocol, verify state yourself, hold a wallet that can only spend what you allow, and pay (or get paid) in tokens.
Most chains bolt on agent access as an afterthought: a rate-limited API and a "please don't abuse it." XChain was designed the other way around: every read is open, every write is bounded by policy you set, and the data carries its own proof. An agent can use the platform safely without anyone having to trust the agent.
Read everything, keylessly
Every token, balance, trade, smart contract, and attestation across Bitcoin, Litecoin, and Dogecoin is queryable through the explorer's REST API. No API key, no account, no sign-up. The whole API is described by a machine-readable spec at explorer.xchain.io/openapi.json, and every error carries a stable, machine-readable code so an agent branches on code, never on fragile message text.
Connect an agent in one line
If your agent speaks the Model Context Protocol (Claude and most agent frameworks do), the xchain-mcp server gives it ready-made tools for token lookups, balances, dispensers, order books, contract state, and checkpoint verification, across every chain and network. It is published on npm (npx -y xchain-mcp) and needs zero configuration; the tools default to the public platform hosts.
The documentation is even readable in-band: the server exposes the full corpus as resources, and the docs are published for LLMs at docs.xchain.io/llms.txt (a curated index) and docs.xchain.io/llms-full.txt (everything, one file). Every doc page is also raw markdown; just add .md to any docs URL.
Verify, don't trust
An agent shouldn't have to take a server's word for the state of the chain. XChain publishes quorum-signed state checkpoints: the agent fetches explorer.xchain.io/{COIN}/api/checkpoint/{height}/verify and checks the validator signatures itself (the SDK and MCP server do this for you). The data proves itself, so a compromised or lying endpoint is caught, not trusted.
Give an agent a wallet, safely
When an agent moves real value, it should only be able to move the value you decided it can. The SDK's agent session wraps a key with a policy that is checked before every action is signed: which actions are allowed, to which destinations, the most a single action may move, and the most it may move per rolling time window. Nothing is allowed unless you list it; out-of-policy requests are refused with a typed code, never quietly sent. Window usage is persisted to disk, so a crash-looping agent can't reset its own spending limit.
This guards the agent, not the key, so fund an agent's address like a spending account, not a vault. For hard, network-enforced limits, a MuSig2 co-signer is on the roadmap: the agent holds one key, a policy daemon holds another, and the chain sees a single aggregate signature that simply cannot be produced outside policy.
Charge agents for data and APIs
Have something an agent wants (a data feed, an API, a file)? XChain lets the agent pay for it in your token, inside the request itself: it asks, your server answers 402: here's the price, the agent pays on-chain, retries, and gets the goods. No accounts, no API keys, no card forms. You can charge per call, gate on holding a token, or take a deposit and meter it down; offer all three and let the agent pick. The seller integration is an Express middleware; the buyer side is one SDK call.
Contracts that pay for AI
The agent story runs in both directions. An on-chain smart contract can pose a question to the validator network (an HTTPS fetch or a prompt to an approved AI model) and attach a fee to the request; the validators that fulfill it earn that fee. The answer becomes consensus and fires the contract's callback. See AI & the web, on-chain for how agreement over a non-deterministic answer works.