AI agents can act on-chain — through MCP

We have shipped neurochain-mcp, a Model Context Protocol server that gives Claude and other AI agents direct access to the network: balances, transactions, validators, governance proposals and contract analysis — and, with a granted session key, real signed actions.

What makes this different

Most chains approach agent access through a trusted middleman application that holds the keys and decides what an agent may do. We do not.

Session-key authorisation is enforced at the consensus layer, inside apply_tx, on every validator. A key's scope, spending cap and expiry are network state, not an app's policy. An agent cannot exceed them by talking to a different app, and no single node can be persuaded to let it.

Tested live

We granted an agent a scoped session key and it signed a real transfer. The transaction confirmed on-chain with the expected balance change, while an unauthorised key was rejected by the node.

You can connect it today wherever local MCP servers are supported.