# MCP Shelf: full shelf > A short shelf of MCP servers that are worth installing. Every server has a config checked against the project's own npm or PyPI package, notes on what it needs, and a setup guide for Claude, Cursor, VS Code and ChatGPT. Read-only API and MCP server for agents. Source of truth: https://mcpshelf.co. API: https://mcpshelf.co/api/v1/servers. MCP: https://mcpshelf.co/mcp. ## Desktop Commander Page: https://mcpshelf.co/servers/desktop-commander Category: Developer tools Summary: A terminal, file search and surgical file edits for your assistant. GitHub: https://github.com/wonderwhy-er/DesktopCommanderMCP Stars: 10129 Desktop Commander turns a chat assistant into something closer to a coding agent: it can run shell commands, keep long-running processes going, search across folders and edit files by replacing exact blocks of text instead of rewriting whole files. It runs with your user account, so it can do anything you can do in a terminal. That is the point, and also the reason to add it to a client you watch rather than one that runs unattended. Good for: - Using Claude Desktop as a coding agent without a separate IDE - Running builds, tests and scripts and reading the output - Large refactors where small targeted edits beat full-file rewrites You need: - Node.js 18 or newer - Nothing else: no account or key Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "@wonderwhy-er/desktop-commander@latest" ] } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 25 tools. ## Serena Page: https://mcpshelf.co/servers/serena Category: Developer tools Summary: IDE-style "find symbol" and "edit symbol" for an agent, backed by language servers. GitHub: https://github.com/oraios/serena Stars: 30730 Serena starts language servers for your project and exposes what they know: where a function is defined, who calls it, and the exact range of a class or method. The assistant can then read and edit by symbol instead of grepping and reading whole files. On big repositories that saves a lot of context, and it makes edits land in the right place more often. Good for: - Large codebases where reading whole files wastes context - Refactors that need every caller of a function - Pairing with a client that already has file access but no code intelligence You need: - uv (for uvx) - Language support for your project language (Serena downloads most servers itself) Config (under mcpServers): ```json { "command": "uvx", "args": [ "--from", "git+https://github.com/oraios/serena", "serena", "start-mcp-server" ] } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 3 tools. Package note: The npm package called "serena" has no link to this project. Serena publishes to PyPI as serena-agent and documents running it from the GitHub source with uvx, which is what the config here does. ## Context7 Page: https://mcpshelf.co/servers/context7 Category: Developer tools Summary: Pulls current, version-specific library docs into the conversation. GitHub: https://github.com/upstash/context7-mcp Stars: 63713 Models learn a library at one moment in time and then keep using that version. Context7 looks up the library you are working with and returns current documentation and examples, so the code you get matches the version you have installed. It has two tools: one turns a library name into a Context7 ID, the other fetches docs for that ID, optionally narrowed to a topic. Good for: - Fast-moving frameworks (Next.js, Tailwind, the AI SDKs) where training data is stale - Asking for code against a specific version - Cutting down on invented function names You need: - Node.js 18 or newer for the local server - A free Context7 API key for higher rate limits (optional) Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "@upstash/context7-mcp" ] } ``` Remote URL: https://mcp.context7.com/mcp Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 2 tools. ## GitHub Page: https://mcpshelf.co/servers/github Category: Developer tools Summary: GitHub's own server for repos, issues, pull requests and Actions. GitHub: https://github.com/github/github-mcp-server Stars: 33976 This is the server GitHub maintains. Through it an assistant can browse code, open and comment on issues and pull requests, check Actions runs and search across repositories you can see. GitHub hosts it, so most people never install anything: point the client at the URL and authenticate. A Docker image exists for anyone who needs to run it themselves. Good for: - Triage: summarise new issues and failing runs - Reviewing or drafting pull requests from the chat - Searching code across an organisation You need: - A GitHub account - A fine-grained personal access token, or OAuth in clients that support it Config (under mcpServers): ```json { "url": "https://api.githubcopilot.com/mcp/", "headers": { "Authorization": "Bearer YOUR_API_KEY" } } ``` Note: Swap YOUR_API_KEY for a fine-grained token limited to the repositories and permissions you actually want the assistant to use. Remote URL: https://api.githubcopilot.com/mcp/ Install check: Needs your setup first. It needs something our Linux test machine does not have (an account, a desktop app, a Mac or a running service), so on 26 Oct 2026 we listed its tools from its package instead of running this config. ## MCP Inspector Page: https://mcpshelf.co/servers/mcp-inspector Category: Developer tools Summary: The official debugging UI for MCP servers. Not a server you add to a client. GitHub: https://github.com/modelcontextprotocol/inspector Stars: 11154 Inspector is maintained by the Model Context Protocol project. It starts a small web app that connects to one server, lists its tools, prompts and resources, and lets you call each tool with your own arguments while showing the JSON going back and forth. Use it before blaming your client: if a server misbehaves in Inspector, the problem is the server or its config. Good for: - Checking a server starts and lists the tools you expect - Debugging environment variables and arguments - Building your own server You need: - Node.js 22 or newer Setup: 1. Run it with the server command after it: npx @modelcontextprotocol/inspector npx -y @playwright/mcp@latest 2. Open the local address it prints in your browser. 3. Click Connect, then List Tools, and try a tool with your own input. Install check: Not yet tested by us. Written from the project's own documentation. We have not run this config ourselves yet. ## XcodeBuildMCP Page: https://mcpshelf.co/servers/xcodebuildmcp Category: Developer tools Summary: Build, run and test iOS and macOS apps, and drive simulators, from an agent. GitHub: https://github.com/cameroncooke/XcodeBuildMCP Stars: 6509 Agents are good at writing Swift and bad at knowing whether it compiles. XcodeBuildMCP closes that loop: it wraps xcodebuild and the simulator tools so the agent can build a scheme, read the errors, boot a simulator, install the app and capture logs. The project now lives under Sentry on GitHub; the npm package is the same. Good for: - Letting Claude Code or Cursor fix its own build errors - Running a test target after each change - Launching an app in a simulator and reading its logs You need: - A Mac with Xcode installed - Node.js 18 or newer Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "xcodebuildmcp@latest", "mcp" ] } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 24 tools. ## Peekaboo Page: https://mcpshelf.co/servers/peekaboo Category: Developer tools Summary: Screenshots of any Mac app or screen, with optional questions about what is on it. GitHub: https://github.com/steipete/Peekaboo Stars: 5395 Peekaboo gives an agent eyes on macOS. It can capture a single window, an app or the full screen, and it can pass the image to a vision model to answer a question such as "is the error dialog still showing?" It is a CLI as well as an MCP server, written by Peter Steinberger, and it needs the usual macOS Screen Recording permission the first time. Good for: - Checking UI changes an agent just made - Debugging desktop apps where the bug is visual - Giving an agent context from apps without an API You need: - macOS - Screen Recording permission for the app running it - Node.js 18 or newer Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "@steipete/peekaboo-mcp" ] } ``` Install check: Needs your setup first. It needs something our Linux test machine does not have (an account, a desktop app, a Mac or a running service), so on 26 Oct 2026 we listed its tools from its package instead of running this config. ## Kubernetes MCP Server Page: https://mcpshelf.co/servers/kubernetes Category: Developer tools Summary: Pods, logs, events and resources from your current kubeconfig context. GitHub: https://github.com/manusa/kubernetes-mcp-server Stars: 2183 This server talks to the Kubernetes API directly (it is a native Go binary, not a wrapper around kubectl) using whatever context your kubeconfig points at. The assistant can list and describe resources, pull logs, check events and, unless you stop it, create or delete things. Keep it read-only on any cluster you care about. Since version 0.0.68 (October 2026) that setting lives in a small TOML file instead of a --read-only flag, and the old flag stops it from starting. The project lives under the containers organisation on GitHub now. Good for: - Asking "why is this pod restarting?" and getting logs plus events - Exploring an unfamiliar cluster - Routine checks on dev and staging clusters You need: - A working kubeconfig - Node.js 18 or newer (the npm package downloads the binary) Setup: 1. Make a text file, for example ~/.config/kubernetes-mcp/read-only.toml, containing one line: read_only = true 2. Put that file's full path in the config below in place of /path/to/read-only.toml. Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "kubernetes-mcp-server@latest", "--config", "/path/to/read-only.toml" ] } ``` Note: Change the file to read_only = false only for clusters where you are happy for the assistant to change resources. The --read-only flag from older guides no longer works. Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 14 tools. ## IDA Pro MCP Page: https://mcpshelf.co/servers/ida-pro Category: Developer tools Summary: Lets an assistant read and annotate the binary you have open in IDA Pro. GitHub: https://github.com/mrexodia/ida-pro-mcp Stars: 13164 IDA Pro MCP has two halves: a plugin inside IDA that exposes the open database, and an MCP server your client talks to. Together they let the assistant decompile functions, follow cross-references, read strings and write back renames and comments. It needs a licensed copy of IDA Pro with Python, so it is for people already doing reverse engineering, not a way to start. Good for: - First passes over large binaries: naming obvious functions - Explaining decompiled code - Keeping notes inside the IDA database as you go You need: - IDA Pro (not the free version) with IDAPython - Python 3.11 or newer Setup: 1. Install the package with the MCP library it was built for: pip install ida-pro-mcp "mcp<2". Without the pin, pip pulls MCP 2 and the server will not start. 2. Run ida-pro-mcp --install. It installs the IDA plugin and adds the server to the MCP clients it finds. 3. Restart IDA and your MCP client, open a binary, and start the plugin from the Edit → Plugins menu. Install check: Started in our test. We installed it with the steps on this page on 26 Oct 2026: it started and answered with 48 tools. Package note: ida-pro-mcp on PyPI links back to mrexodia/ida-pro-mcp. Version 1.4.0 needs the MCP Python library below version 2. ## Playwright MCP Page: https://mcpshelf.co/servers/playwright Category: Browser and web data Summary: Microsoft's browser automation server: navigate, click, type, read. GitHub: https://github.com/microsoft/playwright-mcp Stars: 38652 Playwright MCP gives the assistant a real Chromium, Firefox or WebKit browser. Instead of screenshots it reads pages through the accessibility tree, which is cheaper on context and more reliable for clicking the right button. It is the default choice for "go to this site and do X" and for checking your own web app after a change. Good for: - Testing your web app the way a user would - Filling in forms and following multi-step flows - Reading pages that need JavaScript to render You need: - Node.js 18 or newer - A browser download on first run (Playwright handles it) Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "@playwright/mcp@latest" ] } ``` Note: Add "--headless" to args if you do not want a browser window to open. Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 25 tools. ## Firecrawl Page: https://mcpshelf.co/servers/firecrawl Category: Browser and web data Summary: Scrape pages, crawl sites and search, returned as clean Markdown. GitHub: https://github.com/mendableai/firecrawl-mcp-server Stars: 7691 Firecrawl's server sends requests to the Firecrawl API, which renders pages, strips navigation and ads, and returns Markdown or structured JSON. It can also map a site's URLs, crawl them, and run a web search and scrape the results in one go. Because the work happens on Firecrawl’s side, it copes with JavaScript-heavy pages without a browser on your machine. Good for: - Research across many pages - Pulling structured data (prices, specs, listings) out of sites - Reading documentation sites into context You need: - A Firecrawl account and API key (there is a free tier) - Node.js 18 or newer Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "firecrawl-mcp" ], "env": { "FIRECRAWL_API_KEY": "fc-your-key" } } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 28 tools. Package note: firecrawl-mcp on npm is published from firecrawl/firecrawl-mcp-server. ## Exa Page: https://mcpshelf.co/servers/exa Category: Browser and web data Summary: Search built for AI: web results, page contents and code examples. GitHub: https://github.com/exa-labs/exa-mcp-server Stars: 5170 Exa is a search engine designed to be called by models. Its server returns results with the page text already attached, so the assistant can answer from sources without a second fetch step. A code-search tool finds working examples from GitHub and documentation. The hosted version needs no install, which makes it one of the easiest search tools to add to ChatGPT. Good for: - Giving an assistant up-to-date search with sources - Finding code examples for an unfamiliar API - Research questions that need several pages read You need: - An Exa API key for the local server Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "exa-mcp-server" ], "env": { "EXA_API_KEY": "your-exa-key" } } ``` Remote URL: https://mcp.exa.ai/mcp Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 2 tools. ## Bright Data Page: https://mcpshelf.co/servers/bright-data Category: Browser and web data Summary: Search and scraping through Bright Data, for sites that block ordinary requests. GitHub: https://github.com/brightdata/brightdata-mcp Stars: 2687 Bright Data sells proxy and unblocking infrastructure, and this server puts it behind MCP tools: search engine results, page scraping as Markdown, and a hosted browser for sites that need interaction. Reach for it when simpler scrapers come back with captchas or empty pages. Good for: - Sites that block datacenter traffic - Collecting search results at volume - Structured data from large marketplaces You need: - A Bright Data account and API token Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "@brightdata/mcp" ], "env": { "API_TOKEN": "your-bright-data-token" } } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 32 tools. Package note: @brightdata/mcp on npm is published from brightdata/brightdata-mcp. ## SearXNG Page: https://mcpshelf.co/servers/searxng Category: Browser and web data Summary: Web search through your own SearXNG instance: no vendor, no API key. GitHub: https://github.com/ihor-sokoliuk/mcp-searxng Stars: 1329 SearXNG is an open-source metasearch engine you host yourself. This server points your assistant at it, so searches go through your instance rather than a commercial search API, and it can fetch a result page as Markdown. If you do not already run SearXNG, a commercial option like Exa is less work. Good for: - Search without sending queries to a search vendor - Local or air-gapped setups - Avoiding per-query costs You need: - A running SearXNG instance with JSON output enabled - Node.js 18 or newer Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "mcp-searxng" ], "env": { "SEARXNG_URL": "http://localhost:8080" } } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 4 tools. ## Postgres MCP Pro Page: https://mcpshelf.co/servers/postgres-mcp-pro Category: Databases and data Summary: SQL, query plans, health checks and index advice for Postgres. GitHub: https://github.com/crystaldba/postgres-mcp Stars: 3427 Most database servers stop at running SQL. Postgres MCP Pro adds the things a DBA would check: explain plans, slow queries from pg_stat_statements, index recommendations, and health checks for vacuum, bloat and connections. It has an access-mode switch. Restricted mode keeps it to read-only transactions with time limits, which is where to start on any database that matters. Good for: - Finding why a query is slow - Index suggestions backed by real workload stats - Safe, read-only exploration of production data You need: - Docker - A Postgres connection string Config (under mcpServers): ```json { "command": "docker", "args": [ "run", "-i", "--rm", "-e", "DATABASE_URI", "crystaldba/postgres-mcp", "--access-mode=restricted" ], "env": { "DATABASE_URI": "postgresql://user:password@localhost:5432/dbname" } } ``` Note: The image rewrites localhost in the connection string so it reaches a database on your own machine. Swap restricted for unrestricted only on a database you can afford to change. Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 9 tools. Package note: The npm package called "postgres-mcp" is a different project (llm-graph/postgres-mcp). Crystal DBA publishes this server to PyPI and Docker Hub. In October 2026 "uvx postgres-mcp" falls back to an old 0.1.0 release that crashes, because 0.3.0 pins a pglast version PyPI no longer has, so we publish the Docker image instead. ## Supabase Page: https://mcpshelf.co/servers/supabase Category: Databases and data Summary: Manage Supabase projects, tables, migrations, logs and Edge Functions. GitHub: https://github.com/supabase-community/supabase-mcp Stars: 2971 Supabase maintains this server. It covers the whole project, not just the database: list and create projects, run SQL, apply migrations, read logs, generate TypeScript types and deploy Edge Functions. The easiest setup is the hosted server, which signs you in with your Supabase account in the browser. You can scope it to one project and to read-only from the URL. Good for: - Building a Supabase app with Cursor or Claude Code - Checking logs and advisories when something breaks - Writing and applying migrations You need: - A Supabase account Config (under mcpServers): ```json { "url": "https://mcp.supabase.com/mcp" } ``` Note: Add ?project_ref=your-project-ref&read_only=true to the URL to lock it to one project and to read-only queries. Remote URL: https://mcp.supabase.com/mcp Install check: Needs your setup first. It needs something our Linux test machine does not have (an account, a desktop app, a Mac or a running service), so on 26 Oct 2026 we listed its tools from its package instead of running this config. Package note: The npm package called "supabase-mcp" comes from a different repository (Cappahccino/SB-MCP). Supabase's own server is the hosted URL above; their local package is @supabase/mcp-server-supabase. ## MCP Toolbox for Databases Page: https://mcpshelf.co/servers/mcp-toolbox-for-databases Category: Databases and data Summary: Google's toolbox for Postgres, MySQL, BigQuery, Spanner and other databases. GitHub: https://github.com/googleapis/genai-toolbox Stars: 16728 Toolbox is a server from Google that ships ready-made tool sets for a long list of databases. Pick one with --prebuilt and pass connection details as environment variables, or write your own tools file to expose only the queries you approve. That second option is the reason to choose it: you can give an assistant five named queries instead of raw SQL. Good for: - Teams on Google Cloud databases - Exposing a fixed set of approved queries - One server for several database types You need: - Node.js 18 or newer - Connection details for your database Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "@toolbox-sdk/server", "--prebuilt=postgres", "--stdio" ], "env": { "POSTGRES_HOST": "localhost", "POSTGRES_PORT": "5432", "POSTGRES_DATABASE": "dbname", "POSTGRES_USER": "user", "POSTGRES_PASSWORD": "password" } } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 29 tools. Package note: @toolbox-sdk/server on npm is published from googleapis/mcp-toolbox. ## Qdrant Page: https://mcpshelf.co/servers/qdrant Category: Databases and data Summary: Store notes and find them by meaning in a Qdrant vector collection. GitHub: https://github.com/qdrant/mcp-server-qdrant Stars: 1554 Two tools: store and find. Text goes in with optional metadata, gets embedded locally, and comes back when a later question is close in meaning. It works as long-term memory for an assistant or as a searchable snippet library. It runs against a local Qdrant, Qdrant Cloud, or an on-disk path with no server at all. Good for: - Long-term memory across conversations - A searchable library of code snippets - Small retrieval setups without writing code You need: - uv (for uvx) - A Qdrant instance, or a local path for embedded storage Config (under mcpServers): ```json { "command": "uvx", "args": [ "mcp-server-qdrant" ], "env": { "QDRANT_URL": "http://localhost:6333", "COLLECTION_NAME": "assistant-memory" } } ``` Install check: Needs your setup first. It needs something our Linux test machine does not have (an account, a desktop app, a Mac or a running service), so on 26 Oct 2026 we read its tools from the source code instead of running this config. Package note: The npm package called "mcp-server-qdrant" comes from a different repository (hlos-ai/mcp-servers). Qdrant publishes this server to PyPI under that name, so it runs with uvx. ## Excel MCP Server Page: https://mcpshelf.co/servers/excel Category: Databases and data Summary: Create, edit, format and chart .xlsx files, no Excel needed. GitHub: https://github.com/haris-musa/excel-mcp-server Stars: 4268 This server works on Excel files directly, so it runs on any machine, including servers without Office. The assistant can read ranges, write values and formulas, format cells, add charts and pivot tables, and manage sheets. It can run over stdio for desktop clients or over HTTP if you want to host it. Good for: - Turning data from a conversation into a real spreadsheet - Cleaning or reshaping existing workbooks - Reports generated on a schedule You need: - uv (for uvx) Config (under mcpServers): ```json { "command": "uvx", "args": [ "excel-mcp-server", "stdio" ] } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 46 tools. Package note: The npm package called "excel-mcp-server" has no link to this project. The project publishes to PyPI under that name, so it runs with uvx. ## Blender MCP Page: https://mcpshelf.co/servers/blender Category: Design, 3D and games Summary: Prompt-driven modelling in a running copy of Blender. GitHub: https://github.com/ahujasid/blender-mcp Stars: 32401 Blender MCP connects an assistant to Blender through a small add-on that listens on a local socket. From there the assistant can add and transform objects, set materials, inspect the scene and run Blender Python, while you watch the viewport. It can also pull assets from Poly Haven and generate models through services like Hyper3D when you enable them. Good for: - Blocking out scenes from a description - Learning Blender by asking for changes - Batch edits that would take many clicks You need: - Blender 3.0 or newer - uv (for uvx) - The addon.py add-on from the project installed in Blender Setup: 1. Download addon.py from the project on GitHub. 2. In Blender: Edit → Preferences → Add-ons → Install, pick addon.py and enable it. 3. In the 3D view sidebar, open the BlenderMCP tab and click Connect. 4. Add the config below to your client and restart it. Config (under mcpServers): ```json { "command": "uvx", "args": [ "blender-mcp" ] } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 9 tools. Package note: The npm package called "blender-mcp" has no link to this project. The real server is blender-mcp on PyPI, published from the same author, so it runs with uvx. ## Framelink Figma MCP Page: https://mcpshelf.co/servers/framelink-figma Category: Design, 3D and games Summary: Feeds Figma layout and styling to coding agents so the UI matches the design. GitHub: https://github.com/GLips/Figma-Context-MCP Stars: 16071 Paste a link to a Figma frame and Framelink fetches it through the Figma API, then trims the response down to the layout, spacing, colours and text a coding agent needs. Less noise than the raw API means better first attempts at the component. It reads designs only; it does not change your Figma files. Good for: - Building components from a Figma frame in Cursor - Matching spacing and type without measuring by hand - Design handoff for small teams You need: - A Figma personal access token - Node.js 18 or newer Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "figma-developer-mcp", "--figma-api-key=YOUR-KEY", "--stdio" ] } ``` Note: Replace YOUR-KEY with a Figma personal access token (Figma → Settings → Security). Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 2 tools. ## Talk to Figma Page: https://mcpshelf.co/servers/talk-to-figma Category: Design, 3D and games Summary: Read and edit Figma documents from an agent, through a Figma plugin. GitHub: https://github.com/sonnylazuardi/cursor-talk-to-figma-mcp Stars: 7071 Where Framelink only reads designs, Talk to Figma can change them: create frames and text, set fills and auto-layout, rename layers and read selections. It works through a Figma plugin that connects to a small WebSocket relay on your machine. That means three parts to run, so expect a few minutes of setup the first time. Good for: - Bulk edits like renaming layers or swapping text - Generating rough layouts to refine by hand - Pulling design data into code You need: - The Figma desktop app - Bun (for the relay) - Node.js 18 or newer Setup: 1. Start the relay: bunx cursor-talk-to-figma-socket 2. Install the Cursor Talk to Figma plugin in Figma and run it; note the channel it shows. 3. Add the config below to your client, then ask the assistant to join that channel. Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "cursor-talk-to-figma-mcp@latest" ] } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 40 tools. ## Godot MCP Page: https://mcpshelf.co/servers/godot Category: Design, 3D and games Summary: Launch, run and debug Godot projects, and read the output. GitHub: https://github.com/Coding-Solo/godot-mcp Stars: 6366 Godot MCP gives an assistant the parts of the Godot workflow that need the engine: open the editor, run the project, capture the debug output, and create or change scenes and nodes. The agent can then fix a script and run again without you switching windows. It finds Godot on the usual install paths; set GODOT_PATH if yours is elsewhere. Good for: - Fixing runtime errors with the actual stack trace - Scaffolding scenes from a description - Learning Godot with an assistant that can run your code You need: - Godot 4 installed - Node.js 18 or newer Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "@coding-solo/godot-mcp" ] } ``` Note: Add "env": { "GODOT_PATH": "/path/to/godot" } if Godot is not in a standard location. Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 14 tools. ## KiCad MCP Page: https://mcpshelf.co/servers/kicad Category: Design, 3D and games Summary: Create boards, place footprints and route traces in KiCad from an assistant. GitHub: https://github.com/mixelpixx/KiCAD-MCP-Server Stars: 2959 This server drives KiCad's Python API so an assistant can work on a PCB: create a project, set the board outline, place and move footprints, add tracks and vias, and run design-rule checks. It runs from a clone of the GitHub repository rather than a published package, and it needs the Python that ships with KiCad. Good for: - Quick board outlines and part placement - Repetitive layout edits - Learning PCB design with a guide that can make changes You need: - KiCad 9 with its bundled Python - Node.js 18 or newer - git Setup: 1. git clone https://github.com/mixelpixx/KiCAD-MCP-Server and cd into it. 2. npm install, then npm run build. 3. Point your client at the built entry point, as in the config below, with your own path. Config (under mcpServers): ```json { "command": "node", "args": [ "/path/to/KiCAD-MCP-Server/dist/index.js" ] } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 244 tools. Package note: The npm package called "kicad-mcp" comes from a different repository (iengphogit/kicad-mcp). This project runs from its own source. ## MiniMax Page: https://mcpshelf.co/servers/minimax Category: Design, 3D and games Summary: Text-to-speech, voice cloning, images and video from MiniMax. GitHub: https://github.com/MiniMax-AI/MiniMax-MCP Stars: 1592 MiniMax's own server wraps its media models: speech in many voices and languages, voice cloning from a sample, image generation and short video clips. Files are saved to a folder you choose. Note the API host: accounts from the global site and from mainland China use different hosts, and a mismatch is the usual reason it fails to authenticate. Good for: - Voice-overs for videos or demos - Generating images or short clips inside a chat - Trying voice cloning You need: - A MiniMax API key - uv (for uvx) Config (under mcpServers): ```json { "command": "uvx", "args": [ "minimax-mcp" ], "env": { "MINIMAX_API_KEY": "your-key", "MINIMAX_API_HOST": "https://api.minimax.io" } } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 8 tools. Package note: The npm package called "minimax-mcp-server" comes from a different repository (mgolec/minimax-integration). MiniMax publishes this server to PyPI as minimax-mcp. ## Notion Page: https://mcpshelf.co/servers/notion Category: Knowledge and work apps Summary: Notion's own server: search, read and update pages and databases. GitHub: https://github.com/makenotion/notion-mcp-server Stars: 4680 Notion maintains this server. The assistant can search your workspace, read pages, query databases and create or update content, limited to the pages your integration can see. Notion also runs a hosted version that signs in with OAuth, which is simpler than creating an integration token. Good for: - Drafting docs straight into Notion - Answering questions from your wiki - Updating task databases from a chat You need: - A Notion integration token, shared with the pages you want it to reach - Node.js 18 or newer for the local server Config (under mcpServers): ```json { "command": "npx", "args": [ "-y", "@notionhq/notion-mcp-server" ], "env": { "NOTION_TOKEN": "ntn_****" } } ``` Note: Replace ntn_**** with your integration token and share the relevant pages with that integration. Remote URL: https://mcp.notion.com/mcp Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 24 tools. ## MCP Atlassian (Jira and Confluence) Page: https://mcpshelf.co/servers/atlassian Category: Knowledge and work apps Summary: Jira issues and Confluence pages, Cloud or self-hosted. GitHub: https://github.com/sooperset/mcp-atlassian Stars: 6049 mcp-atlassian is the most used community server for Atlassian tools. It searches and updates Jira issues with JQL, manages sprints and transitions, and reads and writes Confluence pages, for both Cloud and Server/Data Center. You can set it to read-only, and you can limit which Confluence spaces and Jira projects it sees. Good for: - Turning a conversation into well-formed Jira tickets - Summarising a sprint or an epic - Finding the Confluence page that answers a question You need: - uv (for uvx) - An Atlassian API token (Cloud) or personal access token (Server/Data Center) Config (under mcpServers): ```json { "command": "uvx", "args": [ "mcp-atlassian" ], "env": { "JIRA_URL": "https://your-company.atlassian.net", "JIRA_USERNAME": "you@company.com", "JIRA_API_TOKEN": "your-token", "CONFLUENCE_URL": "https://your-company.atlassian.net/wiki", "CONFLUENCE_USERNAME": "you@company.com", "CONFLUENCE_API_TOKEN": "your-token" } } ``` Note: Add "READ_ONLY_MODE": "true" to env while you try it out. Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 98 tools. Package note: The npm package called "mcp-atlassian" comes from a different repository (Vijay-Duke/mcp-atlassian). This project is a Python package of the same name, so it runs with uvx. ## Basic Memory Page: https://mcpshelf.co/servers/basic-memory Category: Knowledge and work apps Summary: Long-term memory as plain Markdown files you own. GitHub: https://github.com/basicmachines-co/basic-memory Stars: 4359 Basic Memory writes notes as ordinary Markdown in a folder on your machine (by default ~/basic-memory) and indexes the links between them. The assistant can write notes, search them and follow connections in later conversations. Because the notes are just files, you can open them in Obsidian or any editor and edit them yourself. Good for: - Carrying project context between chats - A personal knowledge base the assistant can add to - Keeping memory local instead of in a vendor account You need: - uv (for uvx) Config (under mcpServers): ```json { "command": "uvx", "args": [ "basic-memory", "mcp" ] } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 27 tools. ## LinkedIn MCP Server Page: https://mcpshelf.co/servers/linkedin Category: Knowledge and work apps Summary: Read LinkedIn profiles, companies and job posts with your own login. GitHub: https://github.com/stickerdaniel/linkedin-mcp-server Stars: 4016 This server uses a browser session signed in to your LinkedIn account to read profiles, company pages and job listings, and to search jobs. It is a scraper, not an official LinkedIn API. That has two consequences: it can break when LinkedIn changes its pages, and heavy use can get an account restricted. Use it for occasional lookups. Good for: - Researching a company or person before a call - Job searches with a summary of each posting - Prospect research at small volume You need: - A LinkedIn account - uv (for uvx) - A one-time sign-in (the project README shows the current login step) Config (under mcpServers): ```json { "command": "uvx", "args": [ "linkedin-scraper-mcp" ] } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 20 tools. Package note: The npm package "linkedin-mcp-server" has no link to this project, and the PyPI package of that name belongs to someone else. This project publishes to PyPI as linkedin-scraper-mcp. ## arXiv MCP Server Page: https://mcpshelf.co/servers/arxiv Category: Knowledge and work apps Summary: Search arXiv, download papers and read them in full. GitHub: https://github.com/blazickjp/arxiv-mcp-server Stars: 3235 The server searches arXiv by query, date and category, downloads papers, converts them to text and keeps them in a local folder. The assistant can then read whole papers instead of abstracts, and list what you have already saved. It also ships prompts for structured paper analysis. Good for: - Literature reviews - Reading a paper with an assistant that has the full text - Keeping a local library of papers you have looked at You need: - uv (for uvx) - A folder for downloaded papers Config (under mcpServers): ```json { "command": "uvx", "args": [ "arxiv-mcp-server", "--storage-path", "/path/to/papers" ] } ``` Install check: Started in our test. We ran this exact config on 26 Oct 2026: it started and answered with 19 tools. Package note: The npm package called "arxiv-mcp-server" comes from a different repository (r-uben/arxiv-mcp-server). This project publishes to PyPI under that name, so it runs with uvx.