RemoveClaudeWatermark

Free Text-Cleaning MCP Server

Give your agent the same tools this site runs in the browser. A free, hosted, stateless MCP server exposing three tools over Streamable HTTP — install it in Claude Code, Claude Desktop, Cursor or VS Code with one line, no API key and no sign-up.

claude mcp add --transport http watermark https://removeclaudewatermark.online/api/mcp

Then run /mcp inside Claude Code to confirm it connected.

What is an MCP server?

Model Context Protocol is an open standard that lets an AI assistant call external tools through a uniform interface. Instead of pasting text into a web page and pasting the result back, your agent calls the tool directly, mid-task. An MCP server is just an endpoint that advertises a list of tools and executes them on request — this one speaks JSON-RPC 2.0 over HTTP and holds no session state, so there is nothing to configure beyond the URL.

The three tools

clean_text removes invisible unicode characters — zero-width spaces, joiners, bidirectional marks, variation selectors — and normalises lookalike spaces to a regular space, optionally handling em dashes and smart punctuation. analyze_text lists every non-ASCII code point with its unicode name and count, so an agent can inspect text before deciding to modify it. rewrite_text substitutes synonyms, connectives and contractions to disrupt a statistical text watermark, returning the share of words it altered.

Why an agent wants this

Invisible characters break things silently and agents hit that constantly: a zero-width space inside an identifier produces a compiler error on a line that looks perfect, a non-breaking space in YAML is not valid indentation, and a curly apostrophe in a shell string is not a quote character. Giving the agent a deterministic cleaner means it can fix the class of bug rather than guess at it. analyze_text is the diagnostic half — it turns "this string comparison fails for no reason" into a named code point.

What it will not do

rewrite_text does not guarantee removal of Claude's embedded watermark, and the tool description says so to the calling model. That watermark is a statistical property of the model's token choices; the tool degrades it by replacing those choices and reports the substitution rate, which is the only honest metric available. We would rather your agent relay an accurate limitation than a confident falsehood.

Privacy and cost

Free, no key, no account, no rate-limit gymnastics. The server is stateless and stores nothing: text arrives in a request, is transformed by pure functions, and the response is returned. Nothing is logged or retained. It runs on Cloudflare Workers at the edge, so latency is low wherever your agent is.

Frequently asked questions

+ How do I add this MCP server to Claude Code?

Run: claude mcp add --transport http watermark https://removeclaudewatermark.online/api/mcp — then run /mcp inside Claude Code to confirm it connected. No API key is needed.

+ How do I add it to Claude Desktop?

Open Settings → Developer → Edit Config, add the server to claude_desktop_config.json using npx mcp-remote pointed at the endpoint, then restart Claude Desktop. The exact JSON is on this page.

+ Does it work with Cursor and VS Code?

Yes. Cursor reads ~/.cursor/mcp.json (or .cursor/mcp.json per project) and VS Code reads .vscode/mcp.json. Both take the URL directly as an HTTP server — the config for each is above.

+ Is the MCP server free?

Yes, with no key, account or usage limit. It is stateless and stores nothing you send it.

+ What transport does it use?

Streamable HTTP — JSON-RPC 2.0 over POST. It is stateless, so no session initialisation handshake is required and any MCP-compatible client or a plain curl request works.

More free tools

Related guides