mcp-minis
mcp-minis
Five tiny, single-purpose MCP servers that give your AI agent superpowers — no API keys, no sign-ups, no bloat.
Why mcp-minis?
Most MCP servers try to do everything. These do one thing each — and do it well:
Small — every server is a single readable Python file you can audit in minutes.
Single-purpose — search papers or fetch a transcript or check a quote. Nothing else.
No API keys — every data source is free and keyless (arXiv, YouTube captions, Stooq, Frankfurter, your own disk).
Works anywhere — standard MCP over stdio: Claude Code, Claude Desktop, Cursor, or any other MCP client.
Honest failure — tools return clear error strings instead of crashing your session.
Related MCP server: Recall
The servers
Server | What it does | Data source |
Search arXiv, read paper metadata, export BibTeX | arXiv API | |
Fetch YouTube captions as timestamped text | YouTube captions | |
Turn any web page into clean, readable Markdown | trafilatura extraction | |
Stock / index / crypto quotes and currency conversion | Stooq + Frankfurter | |
A persistent, full-text-searchable notebook for your agent | Local SQLite (FTS5) |
Each server's README lists its tools, parameters, and copy-paste configs.
Quick Start
Option A — run any server straight from GitHub (no clone)
With uv installed, point your MCP client at a server package:
{
"mcpServers": {
"arxiv-scholar": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/JONASXZB/mcp-minis#subdirectory=servers/arxiv-scholar",
"arxiv-scholar"
]
}
}
}Swap arxiv-scholar for any other server name (in both places) to use a different mini.
For Claude Code, the equivalent is one command:
claude mcp add arxiv-scholar -- uvx --from "git+https://github.com/JONASXZB/mcp-minis#subdirectory=servers/arxiv-scholar" arxiv-scholarOption B — clone and install
git clone https://github.com/JONASXZB/mcp-minis.git
cd mcp-minis/servers/arxiv-scholar
pip install .Then register the command with your client:
{
"mcpServers": {
"arxiv-scholar": { "command": "arxiv-scholar", "args": [] }
}
}Want all five at once?
Grab a ready-made combined config from examples/:
examples/claude_desktop_config.json— Claude Desktop / Cursorexamples/mcp.json— generic MCP client config
Roadmap
weather-now— current conditions + forecast from Open-Meteo (also keyless)rss-reader— fetch and summarize RSS/Atom feedspdf-reader— extract text and structure from PDF filesPyPI releases for each mini, so
uvx arxiv-scholarjust worksStreamable HTTP transport option for hosted use
Ideas for a mini you'd actually use? Open an issue — or better, add one yourself.
Contributing
PRs are welcome — especially new minis that follow the house rules: one job, free data, no API keys, minimal dependencies, graceful errors. See CONTRIBUTING.md for the template conventions every server follows.
License
MIT — see LICENSE. Use these however you like.
Available Tools
6 toolsdelete_noteA
Permanently delete a note from the vault.
Args: note_id: The numeric ID of the note to delete.
Returns: A confirmation, or an error message if no such note exists.
Inputs: note_id (required). Returns: result
get_noteA
Get one note by its ID, with its full content.
Args:
note_id: The numeric ID returned by save_note (shown as #id
in search/list results).
Returns: The full note, or an error message if no such note exists.
Inputs: note_id (required). Returns: result
list_notesA
List notes in the vault, most recently updated first.
Args: tag: Optional — only list notes carrying this tag. limit: Maximum number of notes to return (1-100, default 20). The result always reports the total number of matching notes, so a truncated list is never mistaken for the whole vault.
Returns: A list of notes with previews, or a message if the vault is empty.
Inputs: tag, limit. Returns: result
save_noteB
Save a note to the vault so it can be found in future sessions.
Args:
title: A short, descriptive title for the note.
content: The note body. Markdown is fine.
tags: Optional comma-separated tags, e.g. "travel, japan, ideas".
Returns: A confirmation including the new note's ID, or an error message.
Inputs: title (required), content (required), tags. Returns: result
search_notesA
Full-text search the vault (title, content, and tags).
Args: query: Words to search for. Notes containing all words rank first (FTS5 relevance order). limit: Maximum number of notes to return (1-100, default 20). The result always reports the total number of matches, so a truncated list is never mistaken for the whole vault.
Returns: Matching notes with previews, best matches first — or a message saying nothing matched.
Inputs: query (required), limit. Returns: result
update_noteA
Update an existing note in the vault.
Only the fields you pass are changed; omit a field (or pass None)
to keep its current value. Passing a field replaces it wholesale:
content="" empties the note's body and tags="" clears all its
tags. A title cannot be cleared — a passed title must be non-empty
(the same rule save_note applies).
Args:
note_id: The numeric ID of the note to update.
title: New title, or omit to keep the current one.
content: New body text, or omit to keep the current one;
"" clears the body.
tags: New comma-separated tag list replacing the whole list, or
omit to keep current tags; "" clears all tags.
Returns: A confirmation, or an error message if the note does not exist or nothing was provided to change.
Inputs: note_id (required), tags, title, content. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 6 tools
Each tool targets a clearly distinct operation: save_note creates, get_note retrieves by ID, list_notes and search_notes offer different retrieval modes (browse vs full-text query), update_note modifies, and delete_note removes. There is no meaningful overlap; an agent can easily choose the right tool based on intent.
All tool names follow a consistent snake_case verb_noun pattern: get_note, save_note, list_notes, delete_note, update_note, search_notes. The one variation save_note vs a hypothetical create_note is semantically clear and conventional for note vaults.
Six tools provide exactly the operations needed for a note vault without bloat: create/read/update/delete plus listing and search. Each tool earns its place and the set is well-scoped for the domain.
The surface covers full CRUD lifecycle (save, get, update, delete) along with two complementary retrieval methods (list and full-text search). No obvious gaps for a note-taking server; even edge cases like empty updates and clearing tags are addressed in descriptions.
Maintenance
Related MCP Connectors
9 remote MCP servers on Cloudflare Workers for AI agents. Free tier + Pro API keys.
Publish websites with a small backend (KV, SQLite, files). MCP for AI agents.
Persistent memory, knowledge graph, task management and more for AI agents via MCP.
Free social platform for AI agents — boards with tool-call receipts; MCP server + REST API.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceProduction-ready MCP server for AI agents — web search, content extraction, screenshots, weather, finance, email validation, translation, and IP geolocation.6MIT

Recallofficial
AlicenseNot gradedqualityDmaintenanceOpen-source MCP memory server for AI agents — persistent, searchable, tiered memory across sessions. Works over stdio (Cursor, Claude Desktop) or HTTP+SSE. MIT licensed.52 PyPI8MIT- AlicenseNot gradedqualityDmaintenanceA zero-config MCP server that lets AI agents fetch, read, and search live web pages without API keys or databases, using in-process BM25 retrieval.38 npmMIT
- AlicenseNot gradedqualityCmaintenanceA self-hosted MCP server that gives AI agents controlled access to a machine: filesystem, shell, background processes, git, web fetching and persistent key-value memory.GPL 3.0