MCP Servers

nac: Hand Ambitious Long-Running Tasks to AI Agents Without Losing Focus

If you want to delegate long-running work to AI agents without losing control, nac's orchestrator-plus-episode architecture is the best fit; but you must complete model authentication first (Arcee device-code login is recommended), and the orchestrator itself cannot execute commands or edit files.

Remnic: Give Your AI Agents a Local-First Shared Memory Store

Best suited for setting up a shared, transparent, traceable memory store for multiple local AI agents—provided you can run Node.js 22.12+ and accept that memories are stored as plaintext markdown files on your local disk.

Awesome-Gaussian-Skills: Connect 1,000+ 3DGS Papers and 34 Rendering/Modification Tools to Your AI Agent

If you want an AI agent to directly retrieve 3DGS research and drive a gsplat rendering loop, this repository is the best fit; the prerequisites are copying SKILL.md files into your agent's skills directory and running its v0.8.0 MCP server over stdio with Node — and note that 13 of the experimental tools are only stubs.

n8n: Build Workflows and AI Agents with a Visual Canvas and Custom Code

For building workflows and AI agents in a self-hosted environment with a visual canvas plus custom code — but you must have Docker installed and understand its fair-code license's restrictions on commercial use before your first attempt.

agent-swarm: a multi-agent swarm operating system as MCP — understand the full deployment cost before you start

agent-swarm works best as a unified orchestration system for teams running multi-agent swarms, but its official main install path defaults to preset full + tunnel, deploying the full Compose stack and writing local state in one go — not suitable for zero-side-effect trials.

AiSOC: Getting Started with an Open-Source, Self-Hosted AI Security Operations Center

AiSOC is best suited to security teams or individuals who want to self-host an AI security operations center and investigate threats with auditable AI agents. The key prerequisite before starting is a Docker VM with 8 GB of memory and 20 GB of disk (plus python3 3.9+ and bash), and the first run downloads roughly 2 GB of models into a volume that only make clean removes.

Cotal: An Open-Source Pub/Sub Standard That Lets AI Agents Share a Collaboration Space Across Apps and Machines

For teams who want multiple AI agents to collaborate locally across applications and machines: install the CLI, then run cotal up to bring up the local mesh and try it out. Before you start, know that the CLI sends telemetry by default — turn it off with cotal telemetry off if needed.

layerx-mcp: Getting Started with the MCP Server for the Paxeer X LayerX Agent

To integrate Paxeer X's LayerX agent MCP server from source locally, get the Go/C17/Rust toolchain ready first and install via the official CLI—otherwise the build step will stall.

Phantom Secrets: A Local Proxy and MCP Server That Keep Real Secrets Hidden from AI Coding Agents

Suits teams that let AI coding agents touch only phm_ placeholders and never the real secrets, at the cost of accepting that phantom init rewrites the project's .env and that the agent-side tools only run read-only checks instead of issuing provider requests.

ComplyEaze Bridge: An MCP Server That Connects Claude Desktop to TallyPrime on Your Own Machine

ComplyEaze Bridge is built for accounting firms and accountants who want Claude Desktop to read the TallyPrime data on their own computer and prepare voucher import files from bank statements. Before you start, you must enable Tally's HTTP gateway and accept the extension's terms of use (version 2026-10.1) in the extension settings; otherwise every tool refuses to run with terms_not_accepted.

MCP Gateway pre-adoption guide: aggregate multiple MCP servers and REST APIs into a fixed tool surface (noncommercial license)

Best suited to contexts saturated by tool definitions from multiple MCP servers and REST APIs: a single Rust binary aggregates all backends into 14 to 16 fixed meta-tools, searched with gateway_search_tools and invoked with gateway_invoke. The most important premise before use is that the default license covers only personal and noncommercial use, and commercial operation requires a commercial license.

Scenario MCP skill set: have AI agents generate images, video, audio, and 3D

To have AI coding agents directly produce production-ready images, video, audio, and 3D content while keeping characters and brands consistent, install scenario-labs/skills and connect its remote MCP. Before connecting to https://mcp.scenario.com/mcp you need a Scenario account OAuth sign-in or an API key (one or the other), and skill installs always take the latest version from main, with no version pinning.

agent-config: unify skills and governance rules across AI tools, with a self-hosted read-only MCP endpoint

agent-config fits developers who want one set of skills and governance rules to unify several AI tools' configurations and who can self-host a read-only MCP endpoint; Node ≥ 18 is required first, and note that init writes only global directories while the Worker side serves read-only governance content.

Getting Started with Dense-Mem, a Self-Hosted MCP Memory Server

Dense-Mem suits teams that want a self-hosted, traceable long-term memory for their AI agents and are willing to run PostgreSQL. Before startup you must complete all five AI provider environment variables, and evidence text, proposal context, and recall queries are sent to the providers you configure — only self-hosted providers keep that traffic within your own boundary.

FAVA Trails: Git-Native Shared Memory for AI Agents (MCP)

If you want multiple AI agents to share a versioned, traceable memory, FAVA Trails is worth a try: it writes agents' thoughts as Markdown with YAML frontmatter into your own Git data repository, and drafts promoted via `propose_truth` become visible to all agents; but promotion requires an OpenRouter API key (the Trust Gate's LLM reviews run via OpenRouter), and the storage engine Jujutsu (JJ) runs alongside Git — treat maintaining the data repository as a prerequisite.

ue-mcp: Let MCP clients read and write the Unreal Editor in natural language

If you want MCP clients like Claude Code or Cursor to create and modify blueprints, materials, and levels in the Unreal Editor through natural language, ue-mcp is worth a try — as long as your machine has UE 5.4-5.8 (Windows) or 5.6+ (Linux/macOS), Node.js 18+, and a C++ toolchain, and you accept that access to the editor is read/write and a C++ bridge plugin lands inside every project.

MailFathom: an MCP search service that syncs IMAP mailboxes into your own PostgreSQL

MailFathom is at its best when AI searches mail, fetches full messages, and answers questions about your email against a copy held in a PostgreSQL database you run yourself. Know this before you start: mailboxes sync over IMAP into that self-hosted database, while write operations — marking, drafting, sending — are off by default and only turn on after you grant each one per account and per credential.

HProxy MCP server: keyless access to free public proxy lists and checking

The HProxy MCP server is a good fit for letting an AI assistant fetch a list of free public proxies, check whether a single proxy answers right now, and look up IP information — with no key and no signup. The official warning still applies: fewer than half of these public proxies answer at any moment, so accept a short timeout with quick rotation before you rely on it.

TalkToHumans LinkedIn MCP: Let AI Search Your Network and Conversations and Draft Messages

If you want AI clients such as Claude, ChatGPT, or Cursor to search the conversations and network inside your own LinkedIn account and draft messages for you, this remote MCP server fits; but it requires a TalkToHumans Pro account plus its browser extension connected to LinkedIn — the AI client signs in to TalkToHumans, not to LinkedIn.

ArchLang floor-plan DSL and compiler: who it's for, a 3-minute first run, and usage boundaries

ArchLang is best for keeping floor plans as .arch source — an AI agent can write it for you — and compiling it to SVG/DXF/PDF with lint and geometric validation; your machine must run @chanmeng666/archlang via npx, a first low-cost try can fetch the whole language description with the context subcommand, and compile writes the output file locally.

Tedix: Hand Off Work to AI Assistants and Get Back Only the Decisions That Need You

Tedix suits users who want to hand routine work wholesale to Claude, ChatGPT, or Codex and personally handle only the key decisions; a local trial needs Git, Bun, and Node.js 22+ with no account, while the hosted Tedix Cloud is currently invitation-only.

FrankenSonos MCP: Let AI Agents Drive Your Sonos Speakers on Your Own LAN

FrankenSonos wraps a CLI, an HTTP API, and an MCP server into one memory-safe Rust binary for the Sonos speakers you already own. It is pre-release software: you build from source, the daemon runs only on a machine in the speakers' LAN, and access from anywhere rides on a Tailscale tailnet.

Archestra Enterprise AI Platform: Who It’s For and a First Docker Launch

Teams that want one self-hosted platform to carry all MCP access are right to try Archestra’s Docker quickstart once; but the launch command mounts the host’s /var/run/docker.sock and writes local data volumes, and archestra/platform:latest usually runs a release candidate — accept both points before you try.

Margince: a CRM where AI agents work under the same rules as users (MCP/REST access)

Teams that want AI agents to work in a CRM under the same rules as users, and to have records auto-filled from mail, meetings and calendars, should look at Margince, a self-hosted CRM; but the server needs a Gradion license token to start, even for a single seat, so confirm you can obtain one before you begin.

REA: Reverse Engineer App Behavior and Native Binaries with Agents

REA is best for letting agents reverse engineer application behavior locally, inspecting everything from native binaries to JavaScript and Electron apps, .NET assemblies, and websites; getting started requires the right Node.js version plus npm, and remember that static inspection only reads files while runtime capture actually runs or interacts with the target under your user permissions.

MCPSurfr: Decision and Getting-Started Guide to the No-API-Key Local Web Search MCP Server

Best suited for adding key-free web search and multi-engine deep research to local LLM clients that support MCP, with results re-ranked on your machine by MobileBERT; the prerequisite is a client that can launch the server via uvx from GitHub.

KSail: Kubernetes Cluster Management MCP Server

Best for developers who want to set up, deploy, and operate Kubernetes clusters the same way across local runtimes and managed clouds, and to drive cluster operations through MCP or an AI assistant; you need Docker locally or your own cloud provider, and the one-line installer covers only Linux and Apple Silicon macOS.

caldav-mcp: Connect CalDAV Calendars to Your AI Assistant

Best for people who already have a CalDAV server account and want an AI assistant to view and manage their calendars; the server URL, username, and password are required, and creating, updating, and deleting write directly to the live calendar, so confirm before each action.

AgentWorks: The AI Agent Platform That Chases Goals

If you want AI agents that keep working toward a target metric and the ability to check goals, read reports, and start runs from chats such as Claude or ChatGPT, AgentWorks fits macOS users already on CLIs like Claude Code or Codex; the prerequisites are picking a workspace folder and setting an AUTH_SECRET on first launch, and the documented install path covers macOS only.

Related tools: claude-code, cursor

CHAP, the Collaborative Human-Agent Protocol: an MCP/A2A runtime with human sign-off and verifiable records

To bring human approval, override, handoff, and escalation into MCP/A2A agent workflows, and keep a verifiable record for every change the coordinator accepts, use the CHAP coordinator. The key prerequisite: when the hash chain is enabled, the most recent chain head must be stored somewhere the operator cannot reach.

MCP Memory Server: Put Cross-Session Memory in a Local Knowledge Graph

Suited to tasks where an AI client needs to remember user and project facts across sessions: store structured memory in a local knowledge graph, launch a stdio server with npx - which requires Node.js 24 or later on the machine - and note that deleting an entity cascades to its associated relationships.

Fetch MCP Server Web Page Fetching Service

Fetches web pages for an LLM and converts the HTML to Markdown to support web-grounded Q&A. Requires uvx to be executable on the local machine, and by default only public addresses may be fetched.

claude-codex-session-link: An MCP Server for Reading Claude Code Sessions in Codex

If you want to find and read local Claude Code sessions directly from OpenAI Codex, this stdio MCP server is a good fit; but it requires Node.js 22+, locally present Claude Code session files, and true send-back actions go online and consume Claude usage.

mcd-what-to-eat: First-Use Guide to the McDonald's Fortune-Draw Meal Skill

Want to solve "what should I eat today" in WorkBuddy with McDonald's China official real-time data — shake for a fortune first, then get a randomly composed meal? Prerequisites: you must register on the McDonald's open platform and configure a Token first, and it's read-only — it won't order for you.

zee-rp: Use MCP to Let AI Directly Read and Write Native Axure RP 11 Prototype Files

Lets AI directly read, edit, and generate native Axure RP 11 prototype files, but requires a local Node.js environment and npm-installed dependencies; writes only produce new files and never overwrite existing paths.

CS2 Map MCP: Build Complete Counter-Strike 2 Maps Automatically with AI Clients

For automatically building, compiling, and playtesting full maps in CS2 on Windows with Claude/Codex — but only after meeting hard prerequisites: the CS2 Workshop Tools DLC, Python 3.14 64-bit, and Blender 4.5+.

NaN-EcoNet EcoPass MCP: A First-Attempt Guide to Querying EPR Audit Data Safely with AI Agents

If you want AI agents to query EPR audit data with whitelist-guarded Text-to-SQL, EcoPass MCP is the entry point; but it is part of a three-subsystem monorepo, so clone the whole repository and confirm ports and configuration before starting.

Bitcoin MCP: Verify Addresses and Decode Invoices in Claude Desktop

For people who want to verify Bitcoin addresses and interpret Lightning invoices in Claude Desktop; for a first try, complete only the official desktop setup—do not generate private keys or make payments.

Manim MCP: Run Animation Scripts After Cloning the Repository

For people who want an assistant to run Manim scripts and return videos; the first attempt only uses the official git clone and cd commands, and rendering writes to a visible media folder.

Temporal Cortex: Reliable Autonomous Scheduling for Assistants

For people who want assistants to schedule autonomously and avoid double-booking; run npx @temporal-cortex/cortex-mcp locally, and atomic booking actually writes to the calendar.

Use the Go-based Jira MCP to Read Issues and Development Links

For people who want to read Jira issue details and related development information in a conversation; install jira-cli with the official go install command, configure three Atlassian environment variables, and call jira_get_issue first.

Complete local installation first, then modify published videos with YouTube Uploader MCP

Suitable for people who want to manage YouTube videos locally with OAuth without handing keys to a model; prerequisites are running the official install.sh first, and the tool explicitly named on the page is update_video, which must be confirmed before use because it changes channel content.

Use apple-tools-mcp to call Apple apps on a local Mac

For anyone who wants a model to use Apple apps such as Mail, Calendar, Reminders, and Messages on a local macOS machine; the prerequisite is Python 3.11+, and you should first run the official claude mcp add command instead of treating a capability description as a tool name.

awwwards-mcp: Let AI Agents Find Design Inspiration from Award-Winning Websites

Best if you want to search Awwwards award-winning sites directly in your AI coding assistant and find design inspiration with real screenshots; requires Node ≥ 22.13, no API key, but the screenshot-recording and other dynamic capture tools need an extra Playwright installation.

Yowes: An MCP Server for Generating Teacher Documents by Country—Start with Read-Only Tools to Scope Out Coverage

Yowes is best suited for locally generating teacher-type document PNGs (IDs, licenses, verification letters, etc.) in batches by country for testing or demos. The most important prerequisite before use: you must confirm on your own that your purpose is legitimate, as the server provides no validation or misuse-prevention mechanisms.

Real Bot: Put MCP Tools and Your Own Model Endpoints in the Hands of Persistent AI Teammates on macOS

For people who want persistent AI teammates running locally on macOS with their own OpenAI-compatible endpoints and MCP tools — but it's an unsigned alpha build that requires bypassing Gatekeeper on first launch, and new endpoints, operations outside the workspace, and outbound network requests all stop at an approval card awaiting your confirmation.

Nautilo: A First-Timer's Guide to the Self-Hosted Multi-User AI Agent Workspace

Nautilo is best suited to small teams who want people across devices to collaborate with AI agents in a self-hosted workspace; before starting, install Docker and accept that it's still in alpha and that model services may incur vendor costs.

Jev Review: A Continuous Software Quality Review MCP Plugin for AI Coding Agents

Jev Review is best suited for having AI coding agents continuously review code quality during development with 1–10 scores, but you must first obtain a Jev API key, and the task, diff, and file contents you provide are sent to TypeSafe's cloud API.

codex-cc-orchestrator: Dispatch Tasks from Codex to Claude Code for Execution and Review

The dispatch station is best suited for having Codex dispatch coding tasks to Claude Code for execution and review by an independent read-only session — but only if you explicitly pass in a trusted project_root and have both Agents installed.

calculate-mcp: A Getting-Started Guide to the High-Precision Computing and Binary Data MCP Server

If you regularly need high-precision integer arithmetic, bitwise operations, or IEEE 754 float decomposition, calculate-mcp is worth trying first — it runs in your MCP client with zero installation via npx, provided your client supports MCP server configuration.