An MCP server, so agents can clean up a wallet too
The Model Context Protocol (MCP) is how an AI agent — Claude, or any other MCP-compatible assistant — calls a real tool instead of guessing. A BetterClaimSol MCP server is planned on top of the same public API, so an agent managing someone's wallet can check its reclaimable rent and prepare the transaction itself.
Not published yet. This page describes what the server is being built to do — watch @betterclaimsol for the release.
What an agent will be able to do
Check a wallet, in conversation
Ask an agent "how much rent can I reclaim?" and have it call the same scan the public scanner runs, then explain the result in plain language — no separate app to open.
Prepare, never execute
The server can return an unsigned transaction for the agent to hand back. It cannot sign or submit anything on its own — every close and every burn still needs the wallet's own signature.
Why this still can't touch your keys
An MCP tool call is a request for data or an unsigned transaction, nothing more. Signing still happens exactly where it always has — in the wallet, with the person present — so an agent can prepare a reclaim or a burn, but never carry one out unattended.
Who this is for
Wallet assistants, portfolio agents, and anyone building on an MCP-compatible client who wants "clean up my dust" to be one tool call away instead of a link to another website. The underlying API is what makes this possible — the MCP server is a thin, agent-shaped layer on top of it.