PromptHouse MCP Server: Personal Prompt Library

Overview

PromptHouse MCP Server exposes your personal PromptHouse prompt library over the Model Context Protocol (MCP), letting AI clients programmatically discover and fetch prompts. Instead of copy‑pasting prompts between notes, apps, and models, an MCP-aware client (for example Claude Desktop, Cursor, or other tools supporting MCP) can call the server to list prompts, retrieve prompt content, and populate model inputs directly.

This server can run in a lightweight NPX process or as a globally installed CLI. It supports both a web mode (connecting to prompthouse.app) and a local mode (pointing to a local PromptHouse instance). The result is a reliable, repeatable workflow for prompt reuse, tagging, and discovery that integrates into developer and end‑user AI toolchains.

Features

• Prompt management integration: expose your saved prompts, titles, and tags to MCP clients.
• Prompt retrieval: fetch full prompt content by ID.
• Two connection modes: web (default) and local (localhost).
• Simple startup via npx, global npm install, or direct GitHub invocation.
• Configurable endpoint, timeout, and debug logging via environment variables.
• Small footprint: runs as a CLI process intended to be launched by an MCP client.
• Troubleshooting aids: verbose logs and sample initialization testing.

Installation / Configuration

Recommended quick start (no install):

npx prompthouse-mcp

Install globally:

npm install -g prompthouse-mcp
prompthouse-mcp

Run directly from the repository:

npx github:newtype-01/prompthouse-mcp

Claude Desktop example configuration (add under mcpServers):

{
  "mcpServers": {
    "prompt-house": {
      "command": "npx",
      "args": ["prompthouse-mcp"],
      "env": {
        "PROMPTHOUSE_ACCESS_LINK": "your-access-link-here"
      }
    }
  }
}

Configuration file locations:

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

Getting your access link:

1. Sign in to https://prompthouse.app (Google sign‑in).
2. Click “Set Up MCP” in the UI.
3. Copy the generated access link and set it as PROMPTHOUSE_ACCESS_LINK.

Environment Variables

Examples:

# Web mode (default)
PROMPTHOUSE_ACCESS_LINK=your-link npx prompthouse-mcp

# Local mode against localhost:3001
PROMPTHOUSE_MODE=local PROMPTHOUSE_ACCESS_LINK=your-link npx prompthouse-mcp

# Custom endpoint and debug
PROMPTHOUSE_CUSTOM_URL=https://my.example/api/mcp-link PROMPTHOUSE_DEBUG=true npx prompthouse-mcp

Available Tools

Once connected via MCP, the server exposes a small set of tools (JSON‑RPC methods) useful to clients:

• get_prompt_list
  • Returns a list of prompts with id, title, tags, and short metadata.
  • Example: “Show me all my prompts tagged ‘image-generation’”.
• get_prompt
  • Returns full prompt text and any associated metadata for a given prompt id.
  • Example: “Fetch prompt content for id abc123 and run it with the current chat context”.

Clients use standard MCP JSON‑RPC calls to interact with these methods. The server responds with structured prompt data ready to be injected into a model request.

Use Cases

• In‑context prompt reuse: Claude Desktop asks the MCP server for an optimized email‑drafting prompt and injects it into the model call—no manual copy/paste.
• Image generation workflows: An image tool fetches a detailed image‑prompt template from your library, fills variables (style, subject), then sends it to a diffusion service.
• Team prompt sharing (local mode): Run a local PromptHouse instance and let team tools query a shared prompt library on localhost for standardized interactions.
• Rapid iteration: Developers keep canonical prompts in PromptHouse, call get_prompt_list to discover variants, then fetch and test variants quickly in their model client.
• Automation pipelines: Scripts call get_prompt to fetch prompts dynamically during batch generation jobs.

Concrete example (Claude Desktop):

1. Configure MCP server as shown under Configuration.
2. Inside Claude Desktop, request: “List my prompts for product marketing.”
3. The client calls get_prompt_list, presents matches, then calls get_prompt to load the selected prompt into the conversation.

Troubleshooting

• “Access link required”: Ensure PROMPTHOUSE_ACCESS_LINK is set and valid.
• Timeouts: Try increasing PROMPTHOUSE_TIMEOUT or verify network/local server availability.
• Client not recognizing server: Restart the client after updating its config and confirm Node.js is accessible if the command uses npx.
• Debugging: Enable PROMPTHOUSE_DEBUG=true and inspect logs for request/response details.

For manual testing you can send an initialize JSON‑RPC payload to the CLI process (example uses shell piping):

echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0.0"}}}' \
  | PROMPTHOUSE_ACCESS_LINK=your-link npx prompthouse-mcp

This will verify the MCP handshake and show server responses in debug mode.

Where to find the project

Source: https://github.com/newtype-01/prompthouse-mcp

Tags: developer-tools