M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
W

Witsy

Community
Not a standard server
Desktop AI assistant and universal MCP client for many model providers

This is not a typical connect-and-use MCP server; read the source repository before using it.

How to install ↓ Compare GitHub source repository ↗
Category
Other #179 of 230
Stars
★ 2.0k Popular
Runtime
Node.js · Prebuilt binary
Credentials
Optional API key
License
AGPL-3.0
Last commit
⚠
45FMRS · D

The source presents Witsy as a feature-rich desktop AI assistant and universal MCP client, not as a separately deployable MCP tool server. Its strengths are broad model-provider support, local models, desktop shortcuts, document retrieval, agents, and an HTTP API. Users should evaluate third-party data handling, API costs, local-file access, and MCP tool permissions before use.

Strongest · Documentation 12/20 Weakest · Reliability 6/20

Reliability
6/20
Security and permissions
8/20
Maintenance
11/20
Documentation
12/20
Setup experience
8/20
Why each score
Reliability 6/20
Only the README is available, and the server manifest is not cached; there is no source, real CI test evidence, or verifiable MCP initialization/tool-list handshake. The README describes MCP support and HTTP endpoints, but startup behavior, declared-versus-implemented tools, error handling, and dependency control cannot be statically verified, so only limited happy-path credit is warranted.
Security and permissions 8/20
The README mentions localhost-by-default binding, Electron sandbox/contextIsolation, MCP authorization, and BYOK, but none is supported by source-file evidence. The application can call many external AI/search services and handle API keys, files, audio, and local data; agent webhooks use short tokens without documented authentication, scoping, rate limits, or confirmation for risky actions. No explicit red-line behavior is evidenced, but permissions and data boundaries are insufficiently demonstrated, requiring substantial deductions.
Maintenance 11/20
The repository is not archived and identifies a maintainer, AGPL-3.0 licensing, 2016 stars, and 55 open issues; the README also shows a test-workflow badge. However, there is no evidence here of commit history, release cadence, dependency updates, security-response procedures, or issue responsiveness, so sustained maintenance and governance are only partially established.
Documentation 12/20
The README provides substantial coverage of installation, providers, API-key prerequisites, HTTP endpoints, CLI usage, agent webhooks, and examples. It lacks a reviewable MCP server manifest, complete tool parameters and limits, cost/privacy disclosures, troubleshooting guidance, and test evidence. It also inconsistently documents HTTP ports 8090 and 4321, preventing a near-perfect score.
Setup experience 8/20
Release binaries, Homebrew, and an npm ci/npm start path are provided, along with prerequisites for many providers and Ollama. However, this is a desktop application rather than a clearly documented standalone MCP server, and no mainstream MCP-client configuration, server launch parameters, or handshake verification steps are supplied. Multiple API keys and local dependencies add manual setup, while the port documentation is inconsistent.

Static review · not runListed 2026-08-14

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesRuns commands or codeUses the networkControls a browser

Best for

  • Desktop users who want one interface for many model providers
  • Users who want to run Ollama models locally
  • Users needing desktop shortcuts, document retrieval, or agent features
  • Developers integrating desktop AI actions through a localhost HTTP API

Not for

  • Users seeking a standalone MCP server that provides tools to other MCP clients
  • Users seeking a confirmed client-specific MCP server configuration
  • Users who do not want to configure cloud API keys or install a local model runtime

Required permissions

  • Cloud models, search, generation, and speech services require the relevant API keys
  • Ollama requires a local installation and model files
  • Local-file RAG and the local filesystem plugin require access to selected files
  • Prompt Anywhere, AI commands, and global shortcuts interact with other desktop applications
  • The local HTTP API listens on localhost port 8090 by default, or the next available port
  • CLI installation may require creating system-level symlinks on macOS or Linux

Risks and side effects

  • User-provided API keys are used with configured third-party services; costs and data handling depend on those providers
  • Local files, prompts, and conversations may be sent to the selected cloud models or services
  • The HTTP API is localhost-only by default; exposing it through a reverse proxy requires user-configured authentication
  • MCP servers, Python execution, agents, and local-file features can expand the actions or data accessible to the model
  • The software is licensed under AGPL-3.0

Setup

Before you start

Runtime：Node.js · Prebuilt binary

Download Witsy from GitHub Releases, or install it on macOS with brew install --cask witsy. To build from source, run npm ci and npm start. Cloud model providers require API keys configured in the application. Ollama requires a local Ollama installation and downloaded models.

Check that it works

Launch Witsy, find the HTTP API port in the tray menu, and run curl http://localhost:8090/api/health — a healthy response confirms the app and its API server are running. Add an API key or install Ollama to start chatting with a model.

Troubleshooting

  1. Confirm that Witsy is running and that the selected provider API key is valid
  2. For Ollama, confirm that Ollama is installed, running, and has the required model downloaded
  3. Check the actual HTTP API port; the README states 8090 for the desktop API but 4321 in the CLI section
  4. For /api/complete, confirm that thread is an array of messages and set engine, model, and stream as needed
  5. For MCP features, verify that the target MCP server is configured correctly; the source does not document a configuration format
  6. For local-file RAG, confirm that a document repository exists and that OpenAI embeddings or an Ollama embedding model is configured

Things to try

Once connected, you can ask your AI assistant things like:

  • Summarize this selected text using an AI command shortcut
  • Answer questions about my PDF via a local document repository (RAG)
  • Trigger my agent through its webhook and check the run status
  • List the available AI engines and models

Use cases

Using multiple cloud and local language-model providers from one desktop application
Running MCP servers to extend model capabilities
Chatting with local files and documents through retrieval-augmented generation
Triggering prompts, AI commands, dictation, and read-aloud actions with shortcuts
Integrating the desktop assistant through its local HTTP API

Overview

Witsy is an Electron desktop AI application and universal MCP client. It connects to cloud and local language models and supports running MCP servers. The README also lists document retrieval, speech transcription, image and video generation, agents, prompts, shortcuts, and a local HTTP API. No MCP server manifest or tool list is provided in the source, so this profile does not identify Witsy as a server exposing MCP tools.

Similar servers

BioMCP 75 · B

One binary. One grammar. Evidence from the biomedical sources you already trust.

★ 654 · Tools 11 Compare with this →
Compare with other MCP servers

Source revision c81c7de417a6 Data synced 2026-10-12 Read the FMRS scoring method