Happy slice of mcp pizzaMCP.pizza

mcp-prompts

MCP.Pizza Chef: sparesparrow

The idea is a shared shelf of reusable wording your assistant can browse instead of you pasting the same instructions over and over. The shipped code does not deliver it yet. Five of the seven advertised tools — add, fetch one, edit, delete and fill in a template — were wired up without telling the assistant what to send, so every call comes back as an error. The listing tool runs but silently drops your tag and search filters, and anything added lives only in memory until restart.

Notes
Writing

Use This MCP server To

See the list of prompt files I keep on disk Check how many prompts and templates I have saved Browse the tags used across my saved prompts Get a quick summary of my whole prompt collection

README

mcp-prompts

Simple MCP server for managing AI prompts and agent configurations with direct claude CLI orchestration.

What It Does

  • Stores prompts as JSON files in data/prompts/
  • Exposes MCP tools for querying and managing prompts
  • Provides agent templates for project orchestration
  • Works with Claude Desktop, Cursor, and other MCP clients

Quick Start

1. Install

pnpm install
pnpm build

2. Start MCP Server

pnpm start

3. Configure Claude Desktop

Add to ~/.claude/mcp.json:

{
  "mcpServers": {
    "prompts": {
      "command": "node",
      "args": ["/absolute/path/to/mcp-prompts/dist/mcp-server-standalone.js"],
      "env": {
        "PROMPTS_DIR": "/absolute/path/to/mcp-prompts/data/prompts"
      }
    }
  }
}

4. Use in Claude

Ask Claude:

  • "List all prompts tagged with esp32"
  • "Get the esp32-fft-configuration-guide prompt"
  • "Create a new prompt for Python FastAPI best practices"

Orchestrating Projects

Use the orchestrate script to analyze entire projects:

./scripts/orchestrate-project.sh ~/projects/mia analyze
./scripts/orchestrate-project.sh ~/projects/esp32-bpm-detector review

This automatically:

  1. Detects project type
  2. Loads appropriate main agent
  3. Spawns specialized subagents
  4. Runs comprehensive analysis
  5. Returns structured results

MCP Tools

Tool Description
list_prompts Query prompts with filters (tags, search, category)
get_prompt Retrieve specific prompt with template expansion
create_prompt Add new prompt to repository
update_prompt Modify existing prompt
delete_prompt Remove prompt
apply_template Apply variables to template string
get_stats Repository statistics

Prompts Organization

data/prompts/
├── main-agents/           # 7 project orchestration templates
│   ├── main_agent_python_backend.json
│   ├── main_agent_cpp_backend.json
│   ├── main_agent_android_app.json
│   ├── main_agent_embedded_iot.json
│   ├── main_agent_multiplatform_iot.json
│   └── ...
│
├── subagents/            # 19 specialized analysis agents
│   ├── explorer.json     # Project discovery
│   ├── analyzer.json     # Code analysis
│   ├── diagrammer.json   # Diagram generation
│   ├── solid_analyzer.json # Code quality
│   └── ...
│
├── cognitive/            # 7-layer cognitive architecture
├── esp32/                # Embedded systems patterns
├── mcp-tools/           # MCP usage patterns
└── [domains]/           # Domain-specific knowledge

Architecture

Simple and focused:

┌──────────────────────────────┐
│   MCP Server (stdio)         │
│   ├── list_prompts           │
│   ├── get_prompt             │
│   ├── create_prompt          │
│   └── ...                    │
├──────────────────────────────┤
│   File Storage               │
│   └── data/prompts/*.json    │
└──────────────────────────────┘

Orchestration:
  orchestrate-project.sh
    ↓ loads prompts
    ↓ builds agent config
    ↓ calls claude CLI
  Actual agent execution

Development

pnpm install       # Install dependencies
pnpm build         # Build TypeScript
pnpm dev           # Watch mode
pnpm test          # Run tests
pnpm orchestrate   # Test orchestration

Enterprise Deployment

For enterprise features (AWS, multi-tenant, payments), see archive/aws/README.md.

Most users don't need this - the local MCP server is sufficient.

License

MIT

mcp-prompts FAQ

Does it work as advertised?
Not right now. Five of its seven tools are registered without declaring what information they accept, so the assistant is handed the request wrapper instead of your input and each call fails with an error. This was checked by running the same setup against the toolkit version the project depends on, not just read off the page.
Are the prompts I create actually saved?
No. The description says prompts are stored as files, but the code only reads those files once at startup and keeps everything in memory afterwards. Anything you add, change or delete through your assistant is gone the next time it restarts.
Does filtering by tag work?
No. The listing tool does run, but the tag and search filters are stripped before they ever reach it, so you get the entire collection back every time regardless of what you asked for.
Does it need an access key?
No. It reads prompt files from a folder on your own computer and never contacts an outside service, so there is nothing to sign up for.
Which apps does it work in?
Claude Desktop and Cursor are both documented, using a settings snippet that points at the built program file and at your prompts folder.
How hard is the setup?
This one is for developers. You clone the project, install its dependencies, build it, then put two absolute file paths into your assistant's settings. There is no ready-made installer.
Is the folder layout in the instructions accurate?
Not quite. It describes a main-agents folder of seven templates and a subagents folder of nineteen, and neither folder exists in the project. What ships is twenty-eight prompt files organised under different names.
Can I use this at work?
Check with someone first. The description and package details both say MIT, but no licence file ships with the project, so GitHub lists it as having no licence at all.
Can I use this to keep a reusable library of prompts?
That is the intention, but not yet the reality. Until the tool wiring is corrected you can see what is on disk and get a count, and little more than that.