Redacta is now in Anthropic's MCP Directory.

Redacta has passed Anthropic's review process and is now listed as an official connector — a desktop extension — in the Anthropic MCP Directory. Anyone using Claude Desktop can find it under Customize → Connectors, add it in a click, and start pseudonymising patient data through natural language: "Redact this letter before I share it."

Unlike most connectors, Redacta is not a remote service you connect to over the internet. It runs entirely in a local process on your own computer. There is no account, no API key, no server — and the text you redact, along with the token map that reverses it, never leaves your machine. For clinical and medical-communications work, that distinction is the whole point.

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic, that lets AI assistants like Claude use external tools through a consistent interface. Instead of the model guessing, an MCP server exposes a defined set of tools with structured inputs and outputs that the assistant can call on demand.

A connector is simply a vetted MCP server that Claude can discover and use. When a tool is listed in the MCP Directory, it has been reviewed by Anthropic's team and made available to Claude users as a trusted integration — a layer of assurance on top of the open protocol.

Redacta detects identifiers with deterministic, checksum-validated patterns — NHS numbers (Modulus-11 validated), National Insurance numbers, dates of birth, postcodes, hospital/MRN numbers, emails and phone numbers — plus general PII such as URLs, IPs, payment cards and account numbers. Clinician names are preserved by design; only the patient's data is removed.

The same server works across MCP-compatible clients, including Cursor, so you can build a pseudonymisation step into your own agent workflows too.

Getting into the MCP Directory is not automatic. Submissions are reviewed against a real bar — tested tools, clear safety annotations on every tool, a privacy policy, a support channel, and proper documentation. For users, that review is the value: it signals the connector behaves as described and has been exercised end to end.

For healthcare and medical communications, the bigger signal is where the data goes: nowhere. Redacta processes everything on-device, makes no network calls, and stores nothing. The token map that can reverse a redaction is handed back to you and never sent anywhere — so you can use modern AI on clinical text while the identifiers stay on your own machine. A built-in HIPAA Safe Harbor mode extends the same approach to US de-identification.

Redacta is one engine shipped across many surfaces — an agent skill, this MCP server, a command-line tool, Python and TypeScript libraries, and a FigJam plugin — so the same detection logic is available wherever clinical text is handled. It is free and open source under the MIT-0 license.