mcp-prompts
A prompt template store with search and variable filling, plus a script that runs Claude agents across a whole project.
- 120 ★ on the repository
- Last commit
Extended access
Why this level
- The server reads and edits a JSON prompt directory, and the orchestration script calls the Claude CLI in your project.
- A loaded prompt enters the agent context as an instruction, so read third-party templates before applying them.
Install
- Works in
Manual install
npm install -g @sparesparrow/mcp-promptsThe README command; GitHub Packages asks for a `read:packages` token, npm needs none.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server reads and edits a JSON prompt directory, and the orchestration script calls the Claude CLI in your project.
- A loaded prompt enters the agent context as an instruction, so read third-party templates before applying them.
This is third-party code. Review the repository files before installing.
What it does
The MCP server keeps prompts and agent configurations as JSON files in `data/prompts/` and serves them to any MCP client. Seven tools (`list_prompts`, `get_prompt`, `add_prompt`, `update_prompt`, `delete_prompt`, `apply_template` and `get_stats`) search by tags and text, add entries and fill `{{variables}}` in templates. The package ships ready sets: nine main agents for project types, subagents for architecture, security and test analysis, cognitive layers and domain templates. The `scripts/orchestrate-project.sh` script detects the project type, loads the matching agent and calls the Claude CLI. The same prompts are available through a JSON API in HTTP mode.
Who it is for. Teams and solo developers who keep prompt templates and agent roles in one place and want to reach them from Claude Code, Claude Desktop or Cursor.
Good fit when
- A prompt template with variables is filled on request, for example a firmware path.
- The agent should list every prompt tagged `orchestration`.
- A whole project is analyzed: the script picks a main agent by project type and launches subagents.
- A non-MCP client pulls the same prompts over HTTP from `/v1/prompts`.
Not a fit when
- Prompts must persist when added through `add_prompt`: such entries live in memory and vanish on restart.
- A multi-tenant paid platform is needed: the AWS variant sits in `archive/aws`.
- One instruction file in the project is enough.
Example request
List all prompts tagged orchestration and apply the claudepy-esp32-android-bridge-workflow template with esp32_firmware_path set to ./firmware.MCP
- Transport
- stdio
- Authentication
- not required
Environment variables
PROMPTS_DIR- Directory of JSON prompts instead of the bundled one.
Limitations
Prompts added or changed through `add_prompt`, `update_prompt` and HTTP live in memory and are lost on restart. A permanent prompt is a JSON file in `data/prompts/` with a stable id, added by pull request. Installing from GitHub Packages needs a token with `read:packages` in `~/.npmrc`, otherwise you get `401 Unauthorized`. The npm install needs no token. The orchestration script calls the Claude CLI, so it spends Claude limits and edits files with the agent permissions.
How to disable. Remove the server with `claude mcp remove mcp-prompts` and run `npm uninstall -g @sparesparrow/mcp-prompts`. If you ran HTTP mode, stop the process or container.
FAQ
How do I connect mcp-prompts to Claude Code?
Install the package with `npm install -g @sparesparrow/mcp-prompts`, then run `claude mcp add -s user mcp-prompts -- mcp-prompts`. Check the connection with `claude mcp list`.
Where are prompts stored?
In JSON files under `data/prompts/`. Set your own directory with `PROMPTS_DIR`. Prompts added through the tools live only in memory.
Is there an HTTP interface?
Yes. `pnpm start:http` (or `npx mcp-prompts-http`) starts a JSON API on port 3003 with `/v1/prompts`, `/health` and `/mcp/tools`.
Which MCP tools are available?
`list_prompts`, `get_prompt`, `add_prompt`, `update_prompt`, `delete_prompt`, `apply_template` and `get_stats`. Templates with `{{variables}}` need `apply_template`.
What does the orchestration script do?
`./scripts/orchestrate-project.sh ~/projects/mia analyze` detects the project type, loads the main agent, spawns subagents through the Claude CLI and returns a structured result.
Discussion
No questions or reviews yet.
Related
MCP serversMCP server for prompt templates, step chains and output validation, exported as skills for Claude Code, Cursor and Codex
PromptOps keeps Claude Code, Codex and Gemini CLI side by side by project and gives a lead agent MCP to launch helpers
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent
A CLI that turns any MCP server, OpenAPI spec or GraphQL endpoint into commands at runtime, with no codegen
A universal MCP CLI client from Apify: persistent sessions, OAuth, JSON output for scripts and shell access to any MCP server for agents
Claude Code Prompts (RepoWise)
claude-code-prompts
Independently written prompt templates for AI coding agents: system prompts, tool prompts, delegation, memory and multi-agent coordination
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| mcp-prompts (this card) | no | not required | 120 | |
| prompts.chat | no | 176,985 | an open prompt library with a plugin and CLI; this is a local server with templates and an orchestration script | |
| PromptScript | no | 344 | compiles rules and skills into files for several agents; mcp-prompts serves ready prompts on request |
README in short
The npm package server reads the JSON files directly and speaks stdio. A fuller hexagonal version with storage adapters lives in `src/mcp/mcp-server.ts`, and the old AWS parts moved to an archive.