An agent can read identity on the XRP Ledger. It still cannot sign.
An identity-focused MCP server covering DIDs, credentials and signer lists, which never holds a key and never submits on its own authority.
Give an agent reach without giving it custody.
The obvious way to let a model touch a ledger is to hand it a key, which is also the fastest way to lose one. This server takes the opposite position: it prepares and decodes and verifies, and it never signs. Every transaction it produces is unsigned and has to be authorized somewhere a human controls.
DIDs, credentials, and signer lists.
Resolving and preparing XLS-40 DID documents, listing and verifying XLS-70 credentials, preparing signer-list changes, decoding a signed blob so you can check it against what you meant to sign, and summarizing an account's identity surface in one call.
Published to npm and listed in the official MCP Registry.
It is MIT licensed and public. An adversarial review pass over the finished code found and fixed five bugs the test suite had not caught, two of them serious, which is the part of the story I would want a client to hear.