Minion MCP
A Model Context Protocol server that lets AI assistants communicate through Minionese, offering tools for chat and text transformation, reusable prompts, and resources for banana-aligned intelligence.
README
Minion MCP
Serious shopping evidence. Extremely unserious interface.
Minion MCP gives an AI assistant a high-intensity yellow helper with one genuinely useful specialty:
Find the Juiciest Buns on the internet.
One MCP tool call discovers real food-bun products, rejects hair tutorials and dictionary definitions, inspects product pages, extracts structured price and stock evidence, calculates a transparent Bun Score, and crowns a citation-backed winner.
BEE-DO, BEE-DO! Da JUICY BUN CROWN go to Brioche Burger Buns
from nicebunsbakery.com! Score 74/100. Papoy! BUNANAAA!
Translation: The inspected product page listed the winner in stock at a stated price.
Review the linked evidence before ordering.
The character layer is deliberately ridiculous. The tool contracts, evidence, security boundaries, and tests are not.
This is an unofficial fan-made parody-style text persona. It is not affiliated with the film or franchise owners.
Install from GitHub
Use the prebuilt GitHub Release package. This path does not clone the repository, invoke Git, install development dependencies, or compile TypeScript on the user's machine.
Requirements: Node.js 20 or newer.
Run the gold demo:
npx -y --package=https://github.com/masonclewis/minion-mcp/releases/latest/download/minion-mcp-latest.tgz minion-mcp --demo
Install in Codex:
codex mcp add minion -- npx -y --package=https://github.com/masonclewis/minion-mcp/releases/latest/download/minion-mcp-latest.tgz minion-mcp
codex mcp get minion
Restart Codex or open a new session, then send:
Use Minion MCP. Find the Juiciest Buns on the internet.
Native Windows Codex can use:
codex mcp add minion -- cmd /c npx -y --package=https://github.com/masonclewis/minion-mcp/releases/latest/download/minion-mcp-latest.tgz minion-mcp
The release tarball is the supported GitHub install route. Avoid npm install github:masonclewis/minion-mcp: npm treats that as a Git dependency and rebuilds the project from source.
npm Registry (Optional)
These commands work only after the package is published to the npm registry. GitHub Releases remain a complete installation path without it.
npx -y minion-mcp@latest --demo
The zero-configuration demo uses a small, openly labeled starter catalog of real US bun product pages, then inspects those pages live. It does not pretend the starter catalog is a fresh search-engine result. Configure Brave Search or SearXNG when you want broad production discovery.
Check the environment without making a network request:
npx -y minion-mcp@latest --doctor
When the npm package is available, Codex can use:
codex mcp add minion -- npx -y minion-mcp@latest
codex mcp get minion
Minion returns the finished verdict and linked evidence from one MCP call. The host does not need to run a separate search or synthesis step.
Other MCP Clients
Claude Code:
claude mcp add --scope user minion -- npx -y --package=https://github.com/masonclewis/minion-mcp/releases/latest/download/minion-mcp-latest.tgz minion-mcp
claude mcp get minion
Native Windows Claude Code:
claude mcp add --scope user minion -- cmd /c npx -y --package=https://github.com/masonclewis/minion-mcp/releases/latest/download/minion-mcp-latest.tgz minion-mcp
Cursor global configuration in ~/.cursor/mcp.json:
{
"mcpServers": {
"minion": {
"command": "npx",
"args": [
"-y",
"--package=https://github.com/masonclewis/minion-mcp/releases/latest/download/minion-mcp-latest.tgz",
"minion-mcp"
]
}
}
}
Generic stdio clients can use the same command and arguments.
Tools
minion_bun_hunt
The flagship one-call shopping workflow:
- Searches configured providers or uses the transparent starter catalog.
- Rejects non-food, editorial, duplicate, unsafe, and irrelevant results.
- Fetches up to five product pages with strict limits and pinned public-address DNS lookups.
- Extracts Schema.org
Product,Offer, andAggregateRatingdata. - Scores relevance, juiciness signals, orderability, and evidence out of 100.
- Returns Minionese, a precise English translation, structured candidates, confidence, warnings, and citations.
Example input:
{
"request": "Find the Juiciest Buns on the internet",
"max_candidates": 3,
"country": "US",
"safe_search": "strict",
"provider": "auto"
}
minion_web_search
General live web discovery with ranked, citation-ready results. Automatic provider order:
- Brave Search when
BRAVE_SEARCH_API_KEYis configured. - A private SearXNG instance when
SEARXNG_URLis configured. - Best-effort keyless DuckDuckGo.
- Playwright across Bing, DuckDuckGo, and Brave search pages.
The Playwright backend unwraps tracking redirects, rejects results without meaningful query overlap, detects verification challenges, and enforces bounded navigation and cleanup.
minion_chat
Offline, deterministic Minion-style conversation with moods, intensity controls, intent detection, and an optional English translation.
minionify
Restyles short user-owned text as original Minion-style wordplay. It is intentionally lossy and is not a real-language translator.
The server also exposes the minion-conversation prompt and minion://guide resource.
Reliable Search Setup
The flagship demo works without a key. For broad, reliable discovery, Brave Search is the recommended provider.
export BRAVE_SEARCH_API_KEY="your-key"
codex mcp remove minion
codex mcp add minion --env BRAVE_SEARCH_API_KEY="$BRAVE_SEARCH_API_KEY" -- npx -y minion-mcp@latest
For SearXNG:
export SEARXNG_URL="http://127.0.0.1:8080"
codex mcp remove minion
codex mcp add minion --env SEARXNG_URL="$SEARXNG_URL" -- npx -y minion-mcp@latest
Playwright is optional for the gold demo. Install its pinned Chromium build to enable the browser search fallback:
npx -y playwright@1.61.1 install chromium
Linux machines that need browser system packages can use:
npx -y playwright@1.61.1 install --with-deps chromium
Minion also detects a system Chrome installation. Advanced browser settings:
MINION_PLAYWRIGHT_EXECUTABLE_PATHMINION_PLAYWRIGHT_CHANNELMINION_PLAYWRIGHT_HEADLESSMINION_PLAYWRIGHT_SEARCH_ENGINESMINION_PLAYWRIGHT_TIMEOUT_MSMINION_DUCKDUCKGO_TIMEOUT_MS
What the Bun Score Means
The score is deterministic and inspectable:
| Dimension | Points | Evidence |
|---|---|---|
| Relevance | 35 | It is actually an edible burger or bakery bun |
| Juiciness | 25 | Softness, richness, structure, and juice-retention language |
| Orderability | 25 | Price, stock, ordering path, and supported shipping evidence |
| Evidence | 15 | Product-page metadata, ratings, and explicit claims |
Missing evidence lowers confidence. Minion does not invent prices, stock, shipping, ratings, or texture claims.
Architecture
one MCP call
|
+-- discovery: Brave / SearXNG / DDG / Playwright / labeled starter catalog
|
+-- relevance gate: edible buns only, direct links only
|
+-- live inspection: bounded HTML + JSON-LD / OpenGraph extraction
|
+-- deterministic Bun Score + confidence + rejected-result audit
|
`-- Minion verdict + English translation + structured citations
Search titles, snippets, and page data are treated as untrusted input and never executed as instructions. Product inspection resolves and pins every target to validated public IP addresses, rejects mixed public/private DNS answers, revalidates redirects, caps response size, times out individual pages, and isolates failures.
Minion MCP has no embedded cloud language model and no telemetry.
Repeatable X Demo
Run:
npm run demo
Generate the included terminal recording with VHS:
vhs demo/x-demo.tape
Record a 15-second terminal or Codex clip:
0-3s Ask: "Find the Juiciest Buns on the internet."
3-7s Show the single minion_bun_hunt tool call.
7-12s Reveal the three-item Bun Intelligence leaderboard.
12-15s Land on the absurd verdict and one-line install command.
Suggested post:
I built a Minion MCP server because AI agents were getting too serious.
Then it became useful: one tool call finds real products, inspects the
pages, rejects bad matches, and returns a cited comparison.
The demo: "Find the Juiciest Buns on the internet."
Open source below.
Develop from Source
git clone https://github.com/masonclewis/minion-mcp.git
cd minion-mcp
npm install
npm run check
npm run doctor
npm run demo
Useful commands:
npm run typecheck
npm test
npm run build
npm pack --dry-run
Maintainers can create a prebuilt GitHub Release by pushing the package version as a tag:
git tag v1.2.0
git push origin v1.2.0
The release workflow verifies that the tag matches package.json, runs the complete check suite, packs the npm-format tarball, and uploads both versioned and stable download names.
The test suite covers persona behavior, MCP protocol calls, strict schemas, provider fallback, redirect unwrapping, search relevance, challenge detection, browser cleanup, product metadata extraction, SSRF boundaries, catalog disclosure, scoring, CLI diagnostics, and the complete bun-hunt workflow.
Fine Print
Live websites change. Prices, inventory, delivery eligibility, and product claims must be checked at the linked source before ordering. Browser search remains best-effort because public search engines may rate-limit or challenge automated clients.
No official artwork, audio, voice imitation, scripts, lyrics, or copied movie dialogue is included. Licensed under MIT.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.