Skip to main content
Glama
nerdychefs-mcp

nerdychefs-mcp

nerdychefs-mcp

An MCP server that gives your assistant access to the free prompt library at NerdyChefs.ai. Search for a task, open a pack, or pick a random prompt to try.

It runs over stdio and fetches prompts when you first need them. Data stays in memory for 60 minutes. Requires Node 20 or newer.

Tools

Name

Arguments

What you get

search_prompts

query (2 to 200 characters); optional category, persona, tag, limit (1 to 25, default 10), include_prompt_text (default false)

Ranked prompt summaries, total matches, and links.

get_prompt

id (integer)

One prompt with its full text and links.

random_prompt

Optional category, persona, tag

A random prompt with full text and the pool size.

list_packs

Optional category, query, limit (1 to 120, default 50)

Packs ordered by title, with descriptions and links.

get_pack

id (slug or exact title); optional include_prompts (default true), include_prompt_text (default false)

Pack details, sections, and optional prompts ordered by ID.

list_categories

None

Categories, subcategories, and prompt counts.

list_personas

Optional top (1 to 873, default 50), query

Personas ordered by count, then name.

list_tags

Optional top (1 to 500, default 50), query

Tags ordered by count, then name.

Category, persona, and tag filters use case-insensitive exact matches. Pack queries search titles and descriptions. Persona and tag queries match part of a name. Search results omit prompt text unless requested. Every argument above also carries its own description in the tool schema, so your client can show you the same detail inline.

Two things catch people out, and both fail quietly rather than with an error:

  • Filter on the category NAME, not its id. list_categories returns both, and only Engineering & DevOps matches. The slug form engineering-devops returns an empty result.

  • get_prompt ids are positions in the published prompt file, not ids from the website. Use an id from search_prompts, random_prompt or get_pack in the same session. An id copied from a nerdychefs.ai page, or saved from last week, can name a different prompt.

The resource nerdychefs://pack/{slug} returns the same JSON as get_pack with prompt summaries and no prompt text.

Related MCP server: prompts.chat MCP Server

Installation for Claude Desktop

Add this entry to your MCP configuration, then restart the client:

{
  "mcpServers": {
    "nerdychefs": {
      "command": "npx",
      "args": ["-y", "nerdychefs-mcp"]
    }
  }
}

Installation for Claude Code

claude mcp add nerdychefs -- npx -y nerdychefs-mcp

Installation for Cursor

Add this to your MCP configuration:

{
  "mcpServers": {
    "nerdychefs": {
      "command": "npx",
      "args": ["-y", "nerdychefs-mcp"]
    }
  }
}

Installation for Codex CLI

codex mcp add nerdychefs -- npx -y nerdychefs-mcp

Configuration

NERDYCHEFS_API_BASE defaults to https://www.nerdychefs.ai/api. Set it in the server's environment to change the API path. Remote URLs must use HTTPS on www.nerdychefs.ai. Local HTTP addresses are accepted for fixture tests. Redirects are refused.

Files load on demand and are cached separately for 60 minutes. Failed requests get one retry. A failed tool call names the file that could not load.

Privacy

The only remote host the server talks to is www.nerdychefs.ai. It downloads public JSON files. Searches and filters run locally. No user queries, prompt inputs, or client conversations are sent to the site.

Nothing is logged or sent anywhere else. There is no telemetry, analytics, or runtime disk cache. Local fixture tests use a loopback HTTP server. Returned website links include the fixed utm_source=mcp, utm_medium=tool, and utm_campaign=nerdychefs_mcp parameters.

Attribution

Prompts from NerdyChefs.ai. Prompts are served live from nerdychefs.ai and remain that site's content. Each prompt result includes attribution and links to its prompt and pack pages. The package does not include the prompt corpus.

Development

pnpm install
pnpm build
pnpm test
NERDYCHEFS_LIVE=1 pnpm test

The regular tests use small local fixtures. The live smoke test runs only when NERDYCHEFS_LIVE=1. Start the compiled server with node dist/index.js, or use pnpm dev while editing. Standard output is reserved for MCP messages.

Licence

The server code is MIT licensed. Copyright 2026 Mathieu Kessler. See LICENSE. This licence covers the server code. Prompts remain NerdyChefs.ai content.

Available Tools

8 tools
get_packB

Open a pack by slug or exact title and optionally include its prompts.

Inputs: id (required), include_prompts, include_prompt_text.

get_promptA

Get a prompt by its numeric ID, including the full prompt text.

Inputs: id (required).

list_categoriesA

List the available prompt categories and their subcategories.

Inputs: no named parameters.

list_packsB

Browse prompt packs by category or search their titles and descriptions.

Inputs: limit, query, category.

list_personasC

Find personas to use when filtering prompts, starting with the most common.

Inputs: top, query.

list_tagsA

Find tags to use when filtering prompts, starting with the most common.

Inputs: top, query.

random_promptA

Pick a random prompt to try, optionally narrowed by category, persona, or tag.

Inputs: tag, persona, category.

search_promptsC

Find prompts for a task, with optional category, persona, and tag filters.

Inputs: query (required), tag, limit, persona, +2 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.6/5.0

    Scored across 8 tools

    Disambiguation5/5

    Each tool targets a distinct operation: retrieval by ID, search with filters, random selection, pack browsing, and metadata listing. There is no overlap between get, search, random, or the list variants.

    Naming Consistency4/5

    Almost all tools follow a clear verb_noun pattern (get_, search_, list_, get_), but 'random_prompt' deviates by using an adjective instead of a verb such as 'get_random'. This is a minor inconsistency in an otherwise uniform naming scheme.

    Tool Count5/5

    Eight tools is well-scoped for a prompt discovery and browsing server. Each tool covers a distinct aspect without redundancy, and the count feels neither sparse nor bloated.

    Completeness5/5

    The tool surface fully covers prompt retrieval (by ID, search, random), pack exploration, and reference data (categories, personas, tags). Since the server appears read-only and discovery-focused, there are no obvious missing operations within that stated purpose.

    Maintenance

    ActivitySlowing
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers