MCP server · OAuth ready

Better prompts, right inside your AI tools.

Improve, refine, shorten, and enhance prompts without leaving your chat. One connector — works in Claude Desktop, Claude.ai, Cursor, Claude Code, Codex, and any other MCP-aware client.

Connect nowManage tokens
MCP Server URL
https://mcp.prompt-architects.com/mcp
Quick start

Pick your client

Each guide takes under a minute. OAuth is the default — you sign in with your Prompt Architects account in a browser tab and consent once.

  1. 1
    Open Settings → Connectors in Claude Desktop.
  2. 2
    Click Add custom connectorand paste the server URL below. No token needed — you'll be asked to sign in via your browser.
  3. 3
    Approve access on the consent screen. Done.
Server URL
https://mcp.prompt-architects.com/mcp

Prefer a config file? Drop this into your Claude Desktop mcp.json:

json
{
  "mcpServers": {
    "pa": {
      "url": "https://mcp.prompt-architects.com/mcp"
    }
  }
}
What you get

Four prompt operations

Each is exposed as a tool the model can call autonomously and as a slash command you can invoke directly.

Improve

/mcp__pa__improve

Rewrite a prompt for clarity, structure, and specificity. Best for everyday polish.

Refine

/mcp__pa__refine

Asks 1–3 clarifying questions, then produces a tailored result. Best for vague prompts.

Shorten

/mcp__pa__shorten

Compress a prompt while preserving intent. Best for context-window or character limits.

Enhance

/mcp__pa__enhance

Add structure, role-framing, and depth. Heavier than improve — best for high-stakes prompts.

Authentication

Two ways to connect

OAuth handles itself end-to-end. Personal access tokens give you a long-lived header for headless setups.

OAuth 2.1 + PKCE

Recommended
  • Paste the server URL — no token to copy.
  • Browser sign-in + consent screen.
  • Tokens auto-refresh; no expiry headaches.
  • Revoke any time from your dashboard.

Personal access tokens

Headless / CI
  • Generate a pa_live_… token in Settings → MCP.
  • Pass it as Authorization: Bearer …
  • Long-lived; revoke from the dashboard.
  • Best for terminals, scripts, and shared dev environments.
Privacy & security

What we send and store

  • Your prompt text is sent to Prompt Architects only when you explicitly invoke a tool. The MCP server never reads your chat history.
  • Transformed prompts are stored in your account history so you can revisit them — same as the web app.
  • Tokens (OAuth and PAT) are stored hashed; raw values are shown exactly once at creation. Revoking from the dashboard is immediate.
  • The MCP server is stateless beyond per-session memory. No chat content is logged.

Ready to ship better prompts?

Sign up free in 30 seconds, then connect from any MCP client.

Create accountView setup guides

Need help? Contact support