MCP integration — Claude Desktop, Cursor, and any MCP client

Turn RLB Whisk into a callable skill for your AI chat client. Instead of asking the AI to spellcheck (unreliable, non-deterministic), the AI calls Whisk and gets a correct answer back.

Install

# Recommended — isolated venv
pipx install rlbwhisk-mcp

# Or plain pip
pip install rlbwhisk-mcp

# Or no-install
uvx --from rlbwhisk-mcp rlbwhisk-mcp

Claude Desktop

Edit the config file (create if missing):

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "rlbwhisk": {
      "command": "rlbwhisk-mcp",
      "env": {
        "RLBWHISK_API_KEY": "rlbwhisk_live_..."
      }
    }
  }
}

If rlbwhisk-mcp isn't on your PATH (plain pip install without pipx), use:

{
  "mcpServers": {
    "rlbwhisk": {
      "command": "python",
      "args": ["-m", "rlbwhisk_mcp"],
      "env": {"RLBWHISK_API_KEY": "rlbwhisk_live_..."}
    }
  }
}

Restart Claude Desktop. RLB Whisk appears in the tools drawer (🔧 icon at the bottom).

Cursor

Settings → MCP → Add server, or edit ~/.cursor/mcp.json:

{
  "mcpServers": {
    "rlbwhisk": {
      "command": "rlbwhisk-mcp",
      "env": {"RLBWHISK_API_KEY": "rlbwhisk_live_..."}
    }
  }
}

Other MCP clients

Any client that speaks the stdio MCP transport can use RLB Whisk. Command: rlbwhisk-mcp. Env: RLBWHISK_API_KEY. That's it.

What the AI can now do

spellcheck "Check the spelling of this paragraph." → structured findings, review-friendly. fix_spelling "Fix the typos in this sentence." → corrected string, ready to use. spellcheck_file "Check /Users/me/notes.md." → reads the file, checks it, reports findings. list_languages "Which languages does RLB Whisk support?" → all 49 ISO 639-1 codes.

Try it

Once configured, type any of these into your chat:

"Check the spelling of Ths is a smple sentnce with a few misspeled wrds."
"Fix any typos in this text: ..."
"Spellcheck the file at /absolute/path/to/notes.md."
"What languages can Whisk check?"

The AI recognizes when to call Whisk and shows you the tool call + response inline.

Why route through MCP?

LLMs are impressive but they don't reliably spellcheck. They miss things, hallucinate "corrections" that aren't corrections, and their behavior changes between model versions. MCP lets the AI hand the task to a deterministic service — the same input always produces the same output. The AI decides when to call; RLB Whisk decides what the answer is.

The same pattern applies to other Whisk tools as they ship (grammar, PDF utilities, OCR, unit conversion, etc). Once your MCP config is wired up, every new Whisk tool is a new skill the AI gains automatically.

Source

MIT licensed. Repo: github.com/rlbwhisk/rlbwhisk-mcp