MCP Prompts

Community

Local MCP server for storing and serving reusable AI prompts and agent templates.

How to install ↓ Compare GitHub source repository ↗

Category: Dev Tools #243 of 438
Stars: ★ 118 Popular
Transport: stdio (local process)
Runtime: Node.js
Credentials: No credential needed
Last commit: 4 days ago
Tools: 7

51FMRS · D

README-only review: a straightforward, self-hosted MCP server for prompt CRUD and template expansion, plus a separate shell-script orchestration layer; no license, authentication, or hosted-service details are given beyond what's stated.

Strongest · Setup experience 13/20 Weakest · Maintenance 7/20

Reliability

8/20

Security and permissions

11/20

Maintenance

7/20

Documentation

12/20

Setup experience

13/20

Why each score

Reliability 8/20: The README lists tools (list_prompts, get_prompt, create_prompt, etc.) and a minimal architecture diagram, but there is no verifiable execution evidence: no CI workflow file, no test file contents — only a mention of the 'pnpm test' command name with no indication of coverage. orchestrate-project.sh drives analysis by shelling out to an external claude CLI, and its correctness/error handling is entirely opaque from the source alone. Per the static calibration rule, reliability cannot exceed 12 without verifiable CI/test evidence; given the thinness of what is shown, it is scored lower at 8.
Security and permissions 11/20: The server itself stores prompts as local JSON files, which is relatively low risk, but destructive operations like delete_prompt/update_prompt have no documented confirmation or backup mechanism. More notably, orchestrate-project.sh invokes an external claude CLI to 'analyze/review' arbitrary external project directories and spawn subagents — this implies cross-directory access and process invocation whose permission boundaries, network exposure, and confirmation flow are undisclosed. The repo also contains archived AWS/multi-tenant/payments enterprise features; while the README says most users don't need them, their presence means payment/cloud-credential-handling code paths may exist that cannot be ruled out by static review. No red-line issue was found (no malware, no credential-theft evidence, no fabricated identity or test claims), so the 0-4 red-line does not apply, but least-privilege scoping and data-flow disclosure remain incomplete.
Maintenance 7/20: Repository metadata lists the license as unknown while the README footer claims 'MIT' — this contradiction itself signals unclear governance and makes the true licensing status unverifiable. 70 open issues against only 117 stars is a comparatively high ratio suggesting limited responsiveness, and no commit history, release cadence, or security-response channel is provided. Overall the project appears maintained but with clear gaps in governance and versioning transparency.
Documentation 12/20: The README is well organized with a quick start, Claude Desktop configuration example, a tools table, directory layout, and an architecture diagram, making it approachable for newcomers. However, tool parameters (input schemas, return formats) are not specified, and there is no limits/cost/error-handling/troubleshooting section; enterprise deployment features are only pointed to via an archive link without in-document risk disclosure. The license contradiction (MIT claimed in README vs. unknown in metadata) further undermines documentation trustworthiness.
Setup experience 13/20: Installation steps are simple and direct (pnpm install → pnpm build → pnpm start) with a clear Claude Desktop mcp.json configuration example including paths and environment variables — a low-friction setup on paper. However, no automated tests or CI evidence in the source confirms these steps actually work end-to-end, so per the static calibration rule setup cannot exceed 15 without verifiable execution evidence; scored at 13 accordingly.

Static review · not runListed 2026-08-26

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local files

Best for

Developers maintaining a personal or team library of prompts and agent templates
Users of Claude Desktop/Cursor who want prompt management exposed as an MCP tool
Projects using claude CLI orchestration with specialized subagents

Not for

Teams needing a hosted/multi-tenant prompt service (enterprise AWS features live in a separate archived module)
Users wanting database-backed prompt storage (prompts are plain JSON files)
Non-technical users unfamiliar with configuring MCP client JSON files

Required permissions

Local filesystem read/write access to the configured PROMPTS_DIR
Ability to execute the Node.js server process
The orchestration script additionally invokes the local claude CLI

Risks and side effects

Prompts and agent templates are stored as plain JSON files with no access control described
create_prompt/update_prompt/delete_prompt let a connected client modify or remove local files with no confirmation step described in the README
No authentication mechanism is documented for the stdio server itself
Enterprise/AWS deployment features exist but are only referenced as archived, without detail in the main README

Setup

Before you start

Runtime：Node.js

PROMPTS_DIR required Directory where prompt JSON files are stored; point it at the repo's data/prompts/ path. Nothing external to obtain.

Run pnpm install to install dependencies.
Run pnpm build to compile the TypeScript project.
Run pnpm start to launch the MCP server.
Add an entry to ~/.claude/mcp.json with command set to node, args pointing to the absolute path of the built dist/mcp-server-standalone.js, and env.PROMPTS_DIR set to the absolute path of data/prompts.

claude_desktop_config.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"}}}}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json

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

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal

claude mcp add prompts -e PROMPTS_DIR=/absolute/path/to/mcp-prompts/data/prompts -- node /absolute/path/to/mcp-prompts/dist/mcp-server-standalone.js

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After configuring the client, check that tools like list_prompts, get_prompt and get_stats appear in the client's tool list; calling get_stats and getting repository statistics confirms the connection.

Troubleshooting

Confirm PROMPTS_DIR points to the absolute path of data/prompts
Run pnpm build before pnpm start if dist/mcp-server-standalone.js is missing
Verify the mcp.json entry uses an absolute path to the built server file
Check pnpm test / pnpm dev output for build errors

Things to try

Once connected, you can ask your AI assistant things like:

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

Tools 7

list_prompts read-only

Query stored prompts with filters such as tags, search text, or category.

get_prompt read-only

Retrieve a specific prompt and expand its template.

create_prompt writes

Add a new prompt to the repository.

update_prompt writes

Modify an existing prompt.

delete_prompt destructive

Remove a prompt from the repository.

apply_template read-only

Apply variables to a template string.

get_stats read-only

Return repository statistics.

Use cases

Storing and reusing prompt templates across projects

Querying prompts by tag or category from within an MCP client

Orchestrating multi-agent project analysis via the bundled shell script

Supported clients

Claude Desktop

CursorPartial support

Listed from the project's documentation, not tested by this site.

Overview

MCP Prompts is a Model Context Protocol server for managing, storing, and providing prompts and prompt templates for LLM interactions. It stores prompts as JSON files under data/prompts/ and exposes MCP tools for querying, creating, updating, and deleting them. The project also ships agent templates for project orchestration and an accompanying shell script that drives the local claude CLI to analyze entire projects. It is designed to work with Claude Desktop, Cursor, and other MCP clients.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →

NPM Sentinel MCP Server 79 · B

AI-powered NPM package analysis MCP server

★ 18 · Tools 19 Compare with this →

MartinLoop MCP Server 79 · B

An independent control layer that gives AI coding agents budgets, brakes, and receipts

★ 413 Compare with this →

GitHub MCP Server 76 · B

GitHub's official MCP server for managing repos, issues, PRs, and workflows via natural language

★ 33.7k · Tools 15 Compare with this →

Compare with other MCP servers

Source revision a20a64a9b8d0 Data synced 2026-10-19 Read the FMRS scoring method