Tools that work offline, on your own device — Works with MCP

37 match your filters — 33 of them answered our check today.

Watching this group for 20 days · how we check

This week12 new arrivals

Narrow it down

Developers9 Freelancers1 Students1

Panguard.AI

Live

Open-source malware scanner and runtime guard for AI agents

Domain registered 4 months ago

FreeWorks offlineOpen source

What to know▼

How it works: Vets AI agent skills and MCP servers for malicious behavior before install, scans what's already installed, and blocks hijack attempts at runtime, using 768 open ATR rules.
What's different: Scanned 96,096 published skills and found 751 malicious; rules already merged into Microsoft, Cisco, MISP, and OWASP tooling.
Pricing: Free, MIT licensed, fully on-device.

Watched for 8 days · last checked 2d ago

Details →

OpenChatCut

Live

Open-source, local-first AI video editor with a real multitrack timeline where Codex, Claude, and MCP agents edit alongside you.

FreeWorks offlineOpen source

What to know▼

How it works: AI agents work on a real multitrack timeline locally
What's different: local-first, editable exports, open alternative to ChatCut
Pricing: free (AGPL)

Watched for 6 days · last checked today

Details →

DocImprint

Live

Generates tamper-evident cryptographic proofs of document contents for AI agents, so legal and compliance teams can verify what an agent actually read.

FreeWorks offline

What to know▼

How it works: POST endpoint turns PDFs/URLs into SHA-256 manifests and Merkle citations
What's different: free offline verification, pay-per-call with USDC
Pricing: free offline verify; pay-per-call via x402 USDC
Best for: legal, finance, and compliance teams using AI agents on documents

Watched for 6 days · last checked today

Details →

Backsight

Live

Transcribes short voice notes on-device into pattern dashboards, fully offline; data syncs end-to-end encrypted across devices, and you can plug in a local MCP server for your own AI agents.

Domain registered 1 month ago

Works offlineCan export my data

What to know▼

How it works: You speak a short voice note, it's transcribed on-device, and an offline dashboard surfaces patterns over time — repetition, distribution, spikes, changes, and correlations. Data syncs end-to-end encrypted across devices with keys you control, and can be exported/imported or connected to a local MCP server for your own agents to analyze.
What's different: Transcription and analysis happen entirely on-device with end-to-end encrypted sync, and data can be exported anytime to avoid lock-in.
Best for: People who want to track personal patterns from quick voice notes without their data leaving their control.

Watched for 11 days · last checked 5d ago

Details →

Local firewall binary that sits between AI agents and the network to block secret leaks, prompt injection, SSRF, and MCP tool poisoning, for developers running autonomous agents, producing a signed offline-verifiable receipt instead of blind trust.

No sign-upWorks offlineOpen source

What to know▼

How it works: A single Go binary scans every outbound request from an agent and writes a signed receipt you verify offline against a published key when it blocks something.
Pricing: Apache 2.0, install via brew.
Best for: Developers whose AI agents have shell access, secrets, and an open line to the internet.

Watched for 7 days · last checked 1d ago

Details →

Diffsmith

Live

Offline code-review studio for uncommitted git changes made by AI coding agents; comment on any line and hand notes back to the agent.

Works offlinePay onceNo subscription

What to know▼

How it works: open a repo, comment on diff lines, return notes to the agent via MCP
Pricing: one-time purchase, lifetime access
Best for: developers reviewing AI agent code

Watched for 6 days · last checked today

Details →

Talamus

Live

A local-first memory system for coding/AI agents that stores notes in plain Markdown with citations for every recall, instead of sending agent knowledge to a hosted service.

No sign-upWorks offline

What to know▼

How it works: Local SQLite/FTS5 index, bitemporal history, corrections routed through review; works as Python library, CLI, MCP server, Docker image, Gemini CLI extension.
What's different: The core works without an LLM call.
Pricing: Apache-2.0, no account required.

Watched for 2 days · last checked 1d ago

Details →

SUUR Things

Live

A free, open-source MCP server that lets AI agents safely read and write your Things 3 tasks, plus adds Kanban, matrix and time-blocking views Things 3 lacks.

FreeNo sign-upWorks offlineOpen source

What to know▼

How it works: reads are read-only, writes go only through Things' official URL scheme so agents can't corrupt data
Pricing: free, open source (MIT), no cloud or account
Best for: Things 3 users on macOS who want AI task management without risking their data

Watched for 5 days · last checked today

Details →

Atlas

Live

Gives developers and AI coding assistants deterministic, cited answers about a codebase's symbols, callers, and dependencies without sending code off-device.

Works offlineData stays with you

What to know▼

How it works: Runs locally as a single binary with SQLite, CLI, MCP and HTTP interfaces to index a codebase
What's different: Deterministic, reproducible results with exact file-and-line citations; claims 36x fewer tokens than a graph baseline
Watch out: Code never leaves the machine

Watched for 4 days · last checked today

Details →

OpenMarkdown

Live

A local-first markdown editor that a developer's AI agent can read and co-edit live in the same file, for people who don't want their notes leaving their disk.

No sign-upWorks offlineNo tracking

What to know▼

How it works: A fast, local-first markdown editor that opens any .md file instantly, and lets an AI agent read, write, and co-edit the same file alongside you via a CLI, agent plugin, and MCP.
What's different: No account or telemetry — files never leave your disk.
Best for: People who want their AI agent editing notes alongside them without the files leaving their machine.

Watched for 9 days · last checked 3d ago

Details →

Bothread

Live

A free, open-source local app that lets multiple AI coding agents (Claude Code, Cursor, Codex, etc.) collaborate on one codebase over MCP while the user watches and approves each step.

FreeWorks offlineOpen source

What to know▼

How it works: Runs locally with no API keys or cloud account required; coordinates coding agents over MCP with step-by-step user approval.
Pricing: Free and open source.
Best for: Developers using multiple AI coding agents on the same codebase who want oversight of each step.

Watched for 7 days · last checked 1d ago

Details →

Creative Studio MCP

Live

Local-first tool for freelancers and creative studios to generate client quotes, payment breakdowns, checklists, and email drafts, connectable to Claude Desktop as an MCP server.

Works offline

What to know▼

How it works: Runs in the browser, saves data on the user's computer, and can connect to Claude Desktop as an MCP server.
Best for: Freelancers and creative studios producing client-facing quotes and project documents.

Watched for 11 days · last checked 5d ago

Details →

Engine Mail

Live

An AI email client that unifies Gmail, Outlook, iCloud, Yahoo and IMAP inboxes, runs its AI locally or with your own key, and actively blocks trackers and phishing links.

Works offline

What to know▼

How it works: Unifies Gmail, Outlook, iCloud, Yahoo and other IMAP accounts into one inbox, with AI (run locally or via your own key) that summarizes threads, extracts tasks, and writes replies, plus tracker blocking, phishing detection, link/attachment scanning and one-click unsubscribe.
What's different: AI can run locally or with your own API key instead of a vendor-hosted model, and it works offline.
Best for: People who want an AI email client with built-in security screening across multiple connected inboxes.

Watched for 10 days · last checked 4d ago

Details →

Tilores Studio desktop entity resolution

Live

A desktop entity-resolution tool that matches and deduplicates records in a CSV locally on your machine, free up to 100k records, with an MCP server for AI coding assistants.

FreeNo sign-upWorks offline

What to know▼

How it works: Loads a CSV and resolves duplicate records live using the same real-time matching engine as Tilores' cloud product, running entirely on your machine; a local MCP server also lets Claude Code, Codex and other AI assistants search, import and steer it against your data.
What's different: Skips the usual enterprise entity-resolution sales process (demo, contract) — nothing leaves your laptop, and it's free up to 100k records.
Pricing: Free up to 100,000 records.
Best for: Teams who want to deduplicate and match records locally, including via AI coding assistants, without a cloud enterprise contract.

Watched for 10 days · last checked 4d ago

Details →

Local MCP (LMCP)

Live

A local MCP server that gives AI assistants like Claude, Cursor and ChatGPT tools to act on your Mac — iMessage, Teams, WhatsApp, Mail, files — with nothing sent to the cloud.

FreeWorks offlineData stays with you

What to know▼

How it works: A local Model Context Protocol server running on a Mac that gives AI assistants like Claude, Cursor and ChatGPT 143 tools to act on iMessage, local Microsoft Teams chats, WhatsApp, Apple Mail, Calendar, Contacts, Notes, Reminders, Finder files, OneDrive, Outlook and Office documents.
What's different: Runs 100% locally with no cloud connection, reaching apps like iMessage and local Teams chats that cloud-based connectors can't access.
Pricing: Free.
Best for: Mac users who want their AI assistant to act across native apps like iMessage and Mail without sending data to the cloud.

Watched for 18 days · last checked 2d ago

Details →

Fudge MCP

Live

Lets AI coding agents pull real design references — fonts, colors, layouts — from thousands of live sites instead of guessing, via a local MCP server and Chrome extension.

Works offline

What to know▼

How it works: A design reference engine that lets AI coding agents search nearly 10,000 real websites by fonts, colors, components, layouts and visual similarity, running locally through MCP.
What's different: Grounds an AI agent's design choices in real, measured references and screenshots instead of asking it to invent a 'modern, premium' look from scratch.
Best for: Developers using AI coding agents who want evidence-based design references, with a Chrome extension to save ones they like.

Watched for 14 days · last checked 8d ago

Details →

Mispher

Live

An on-device dictation, rewriting, translation and local AI agent tool for Apple Silicon Macs - free and open-source (MIT), with no cloud, no accounts, and MCP support.

FreeWorks offlineOpen source

What to know▼

How it works: Runs on-device dictation, rewriting, and translation using multiple transcription models (Parakeet, CTC, Nemotron, Qwen3-ASR), plus a local agent that plans and calls tools like Apple Notes, clipboard, files, and MCP.
What's different: Fully local processing with no cloud and no accounts, and lets you swap or customize any underlying model or prompt.
Pricing: Free and open-source (MIT license).
Best for: Apple Silicon Mac users who want dictation and an AI agent without sending data to the cloud.
Watch out: Requires Apple Silicon and macOS 26; some listed models (Gemma 4, Qwen3.6, Ornith 1.0) are described as coming next, not yet available.

Watched for 17 days · last checked 11d ago

Details →

Auxly

Live

An open-source, local-first memory and task layer for AI coding agents, built around MCP. Developer infrastructure, not a consumer product.

Works offlineOpen source

What to know▼

How it works: An open-source, local-first memory layer for AI coding agents, with an MCP-powered task manager for tracking agentic workflows.
What's different: Runs locally rather than storing agent memory in the cloud, and is built specifically around MCP for AI code editors.
Best for: Developers building or using AI coding agents who want persistent, local memory and task tracking instead of ephemeral context.

Watched for 18 days · last checked 11d ago

Details →

Infrawise

Live

An open-source local MCP server that gives AI coding assistants like Claude Code, Cursor and Copilot a live map of your AWS setup, database schemas and infrastructure-as-code.

Works offlineOpen source

What to know▼

How it works: An open-source MCP server that gives AI coding assistants like Claude Code, Cursor, and Copilot a live, read-only map of your AWS services, database schemas, and infrastructure-as-code, including exact partition keys, Lambda event shapes, and misconfiguration warnings.
What's different: One command, zero config, runs 100% locally with no data leaving your machine.
Pricing: Open source.
Best for: Developers using AI coding assistants who want those assistants to understand their real AWS infrastructure instead of guessing.
Watch out: Read-only and local-only by design, but it does connect to your live AWS account to build its map.

Watched for 15 days · last checked 9d ago

Details →

Aurora

Live

Runs real statistical and causal analysis locally instead of letting an LLM guess, citing every result to its source. Built for technical users via its MCP server and Python SDK.

Works offlineOpen source

What to know▼

How it works: Runs 24+ research-grade quantitative methods locally on a dataset you drop in — anomaly detection, causality, regime detection, forecasts — with every finding cited to the method and source behind it. Includes an MCP server and Python SDK for AI agents to call it for verified math.
What's different: Computes real statistics locally instead of having an LLM guess at numbers, with a "0 fabricated" indicator guaranteeing no invented figures.
Pricing: Open source, Apache 2.0.
Best for: Developers and technical users who need verified quantitative analysis, including AI agents that call it programmatically.

Watched for 15 days · last checked 9d ago

Details →

Mother Brain

Live

Gives AI coding tools like Cursor and VS Code persistent local memory of your codebase, running on your own machine - priced at $25/year, not one-time, apart from a limited beta deal.

Works offline

What to know▼

How it works: Gives AI coding assistants persistent memory: it indexes your codebase with vector search, saves every conversation, and connects to Zed, Cursor, VS Code and other MCP-compatible tools, running on embedded PostgreSQL + pgvector.
What's different: 100% local-first with no cloud required, unlike most AI memory tools.
Pricing: $25/year, no tiers; beta testers get lifetime access for a $25 one-time payment.
Best for: Developers using AI coding assistants who want persistent, local memory of their codebase across sessions and team members.

Watched for 15 days · last checked 2d ago

Details →

HyperNexus

Not responding

A local-first control system for coordinating multiple AI agents with persistent memory and tool routing.

Domain registered 1 month ago

Works offline

What to know▼

How it works: A local-first control system for coordinating AI agents, with progressive MCP and skill tool routing, persistent memory, and LLM waterfall failover.

Watched for 6 days · last checked today

Details →

Muster

Live

Desktop app that runs each AI coding-agent task in its own isolated git worktree with DB tooling and PR review, offline-first.

Domain registered 16 days ago

Works offline

What to know▼

How it works: Runs each AI coding-agent task in its own isolated git worktree, with database access through MCP, hunk review, PR creation, SSH, a five-engine DB studio, and Linear/Jira/GitHub imports.
Pricing: $39 one-time per device.
Watch out: Offline-first: code and credentials never leave your machine. Available on macOS, Linux, Windows.

Watched for 3 days · last checked 2d ago

Details →

Prism

Live

A native macOS AI workspace with local-first chat, writing assistance, and browser automation, extensible by developers via an MCP registry.

Works offline

What to know▼

How it works: A native macOS AI workspace with a multi-provider chat interface, local-first data handling, system-wide writing assistance, browser automation, file creation, and study tools that generate linked quizzes and flashcards.
What's different: Developers can extend it via an MCP registry, @-tools, and Prism Hosted capabilities.
Watch out: macOS only.

Watched for 8 days · last checked 2d ago

Details →

Synapsor Runner

Live

A tool that replaces raw SQL access for AI database agents with human-approved, scoped business actions instead of full read/write authority.

No sign-upWorks offline

What to know▼

How it works: tenant-scoped semantic tools like billing.propose_late_fee_waiver; writes need human approval outside the model's reach
What's different: includes a CLI audit command to score an existing MCP setup's risk
Pricing: local-first, open source (Apache-2.0), no account needed

Watched for 5 days · last checked today

Details →

Toolport

Live

A local gateway that lets every AI coding client share one set of MCP server connections instead of configuring each tool separately, cutting how many tools your agent has to load.

FreeNo sign-upWorks offlineOpen source

What to know▼

How it works: A local-first, open-source gateway. You set up and authenticate each MCP server once, and every connected AI client — Claude, Cursor, VS Code, Codex — shares that same connection instead of you configuring it per tool.
What's different: Claims up to 91% less tool-token overhead by giving an agent a handful of meta-tools instead of loading hundreds of individual tool definitions; API keys stay in the OS keychain.
Pricing: Free, open source, no sign-up.
Best for: Developers running multiple AI coding clients against the same set of MCP servers who don't want to re-authenticate each one separately.
Watch out: The '91% overhead cut' figure is the product's own claim; we haven't independently benchmarked it.

Watched for 20 days · last checked today

Details →

MemPlato

Live

Turns your Android phone into a self-hosted MCP server so AI assistants like Claude, Cursor or Perplexity can pull your personal memory without sending data to the cloud. Aimed at developers already using MCP-compatible tools.

Works offlineData stays with youOpen sourceNo subscription

What to know▼

How it works: Turns an Android phone into a self-hosted MCP (Model Context Protocol) server by installing a Python MCP server inside Termux, with a secure tunnel giving the phone a stable public HTTPS URL that AI clients like Claude, Cursor or Perplexity can connect to for persistent memory.
What's different: Data is stored locally on the user's own phone instead of a vendor's cloud, with no subscription or vendor lock-in.
Pricing: No subscription.
Best for: Developers already using MCP-compatible AI tools who want to own their memory data instead of storing it in the cloud.
Watch out: Relies on a relay tunnel (relay.memplato.com) and requires Termux running on the phone, with a watchdog to restart the server if the phone reboots or Termux closes.

Watched for 10 days · last checked 4d ago

Details →

Cometline

Live

A local-first desktop AI assistant for macOS with memory, agent delegation and MCP tool support, working with multiple external AI providers rather than being fully offline.

Works offline

What to know▼

How it works: A local-first desktop AI assistant for macOS with semantic memory, agent delegation, slash-command skills, and MCP tool support, connecting to multiple external AI providers.
What's different: Runs locally with a native-feeling macOS interface while still supporting multiple AI providers and MCP tools, rather than being tied to one vendor.
Best for: macOS users who want a desktop AI assistant with persistent memory and MCP tool access rather than a browser-based chat.

Watched for 11 days · last checked 5d ago

Details →

Atlarix

Live

A desktop coding harness built specifically around open-weight models like DeepSeek, Qwen and Kimi, instead of bolting them onto a Claude or GPT-first interface as an afterthought.

FreeWorks offline

What to know▼

How it works: A desktop AI coding harness built around open-weight models (DeepSeek, Qwen, Kimi, MiniMax) instead of Claude/GPT, using a structural + lexical code-retrieval engine (Blueprint) that builds a real dependency graph of the codebase rather than embeddings or a vector DB.
What's different: Built specifically for open-weight models' weaker context discipline, instead of treating them as an afterthought behind a Claude/GPT-first interface.
Pricing: Free.
Best for: Developers who want to use open-weight models like DeepSeek, Qwen, or Kimi for coding with retrieval built for their limitations.
Watch out: Depends on local tools like Ollama and MCP — offline/no-cloud setup requires configuring those dependencies yourself.

Watched for 18 days · last checked 2d ago

Details →

Vellium

Live

A free, open-source desktop app for AI character chat, roleplay and long-form writing, where your data stays in a local database even with a cloud AI backend.

FreeWorks offlineOpen source

What to know▼

How it works: A desktop AI workspace for character chat, multi-character roleplay, and long-form writing with RAG, LoreBooks, and MCP tools and plugins; you bring your own backend, such as OpenAI-compatible APIs, OpenRouter, LM Studio, Ollama, or KoboldCpp, while chats, characters, projects, and knowledge are stored in a local SQLite database.
What's different: Keeps your data in a local database even when you connect a cloud AI backend, and is free and open source.
Pricing: Free and open source.
Best for: Writers and roleplayers who want to choose their own AI backend while keeping their data local. Native builds are available for macOS, Windows, and Linux.

Watched for 14 days · last checked 2d ago

Details →

MemoryLayer

Live

A local-first, MCP-native memory layer for AI agents that reasons over stored facts, catches contradictions, and verifies claims before an agent repeats them — built for developers.

Works offline

What to know▼

How it works: A local-first, MCP-native memory layer for AI agents that reasons over stored facts, catches contradictions, derives facts that were never explicitly stated, and checks claims before an agent repeats them.
What's different: Positioned as reasoning and formal verification over stored knowledge, not just a vector database for similarity lookup.
Best for: Developers building AI agents who need a memory layer with consistency checks.

Watched for 12 days · last checked 6d ago

Details →

ScienceX

Live

An open-source, self-hostable desktop workbench for scientific research that logs experiment data, agent sessions, and provenance locally.

Works offlineOpen sourceSelf-hosted

What to know▼

How it works: Local-first desktop app for macOS/Windows/Linux storing experiment runs and provenance
What's different: Self-hosted alternative to Claude Science, extensible via Skills and MCP
Pricing: Open source
Best for: Researchers wanting to keep experiment data and computation local

Watched for 6 days · last checked 5d ago

Details →

Wisp

Live

A hotkey-triggered AI overlay for your desktop that reads on-screen context and answers by voice or text via MCP, using a model provider you supply.

Works offline

What to know▼

How it works: A hotkey-triggered AI overlay for the desktop — press a key, pick an intent, and it reads on-screen context, then streams an answer with voice or text without leaving your current app.
What's different: Local-first and lets you bring your own model provider via MCP, instead of locking you into one AI vendor.

Watched for 15 days · last checked 2d ago

Details →

mcpfold

Live

A CLI and proxy that curates your MCP server config and shares it across Claude, Cursor, VS Code and Windsurf, loading only the tools each agent needs — for developers juggling multiple AI coding clients.

Works offline

What to know▼

How it works: A local-first CLI and proxy where you write your MCP server configuration once, then it fans that config out to Claude, Cursor, VS Code, and Windsurf, loading only the tools each agent actually needs and resolving secret references instead of hardcoded keys.
What's different: Loads a curated subset of tools per client rather than every configured MCP server, avoiding the context-window overhead of loading them all.
Best for: Developers using multiple AI coding clients who want one MCP config instead of duplicating setup across each tool.

Watched for 16 days · last checked 2d ago

Details →