Best for
- Individuals or teams who want control over their data and need a self-hosted AI chat workspace
- Developers familiar with Next.js, React, and TypeScript
- Users needing multi-provider integration and plugin extensibility
N
Neo Chat
Community
Not a standard server
A local-first AI chat workspace for models, agents, skills, plugins, search, RAG, voice, memory, and artifacts.
This is not a typical connect-and-use MCP server; read the source repository before using it.
- Category
- Other #88 of 230
- Stars
- ★ 1.9k Very popular
- Runtime
- Node.js 22+ · Docker
- Credentials
- Optional API key
- License
- MIT
- Last commit
58FMRS · C
Neo Chat is a feature-rich local-first AI chat application emphasizing data ownership and self-hosting. It offers multi-provider integration, skills, plugins, RAG, search, voice, and sync, suitable for technical users. However, it is not a turnkey public SaaS and requires secure deployment and configuration.
Strongest · Documentation 14/20 Weakest · Reliability 8/20
Why each score
- Reliability 8/20
- The source code indicates a large Next.js application with extensive testing (unit tests, e2e tests) and CI workflows, but the MCP server functionality (remote MCP support, stdio bridge) is only a part of the larger app, and there are no dedicated tests for MCP handshakes and tool listing. No manifest file is provided, and the source code depth was not verified. Hence, reliability score is limited by static review and lack of direct MCP-specific evidence.
- Security and permissions 12/20
- The app has a well-defined security model: BYOK encryption, local-first storage, access password, confirmation for destructive tool calls, and URL safety gates. The README declares these but they were not directly verified in source. Deployment docs explicitly state that self-hosting is not a turnkey public SaaS security boundary. Deductions are made because implementations were not verified in code and docs acknowledge additional production hardening needed. No red flags (like malware or credential theft) were identified.
- Maintenance 14/20
- The repository shows multiple releases (v2.4.0, v2.2.0 etc.), a public changelog, recent commits (dated 2025 per README), and CI and Docker workflows. License is MIT. While there is evidence of active maintenance, the cadence of releases and issue responses is not visible, and dependency updates are not monitored, so deductions are applied.
- Documentation 14/20
- The README is very comprehensive covering installation, deployment, configuration, architecture, security model, and FAQ. There is a docs directory mentioned (deployment-hardening, reliability-model) and guides for environment variables and plugin development. Although documentation is detailed, it lacks client-side configuration examples for MCP servers (e.g., Claude Desktop config), and the MCP stdio bridge doc is not provided in the material, so deductions are applied.
- Setup experience 10/20
- Installation steps are clear: Node.js 22 and pnpm required, run pnpm install and pnpm dev. Deployment guides exist for Vercel and Cloudflare Workers. However, specific steps to connect as an MCP server in a client are not outlined; the README mentions remote MCP servers and stdio bridge but without direct client configuration examples. Also no manifest file is provided, so deductions are applied.
Static review · not runListed 2026-08-07
Read the FMRS scoring method →Fit and risk
What it can accessReads local filesWrites / deletes local filesRuns commands or codeUses the network
Not for
- Public multi-user services requiring a turnkey SaaS security boundary (accounts, tenant isolation)
- Users who do not want to manage their own deployment and configuration
- Users needing native agent orchestration (only limited Agent mode in browser)
Required permissions
- Browser storage access (localStorage, IndexedDB, OPFS)
- Network access to connect to model providers, search, RAG, voice, and MCP servers
- Server-side API routes execute plugins, MCP, and proxy requests
- Optional server environment variables for deployment defaults and security
- Docker deployments may expose port 3000
Risks and side effects
- Self-hosting entails security responsibilities; public exposure requires additional hardening
- User-entered provider credentials are encrypted in the browser, but server misconfiguration could expose them
- Hosted mode depends on shared stores like Upstash and restricts network policies
- Remote MCP servers can execute tools; configure carefully
- Data is stored in the browser by default; clearing browser data may cause loss unless backed up or synced
Setup
Before you start
Runtime：Node.js 22+ · Docker
ACCESS_PASSWORD
Deployment access gate; not an account system.
BYOK_PRIVATE_KEY_PEM
Stable production BYOK server private key used to encrypt user-entered secrets.
UPSTASH_REDIS_REST_TOKEN
Upstash Redis REST credential for shared short-lived state in hosted or multi-instance deployments.
DEFAULT_PROVIDER_API_KEY
API key for the server-default model provider, obtained from Google/Anthropic/OpenAI etc.
DEFAULT_SEARCH_API_KEY
Optional key for external search providers (Tavily, Firecrawl, etc.) to raise rate limits.
DEFAULT_RAG_TOKEN
Access token for the external vector/RAG service.
DEFAULT_MINERU_API_TOKEN
DEFAULT_LLAMA_PARSE_API_KEY
DEFAULT_ELEVENLABS_API_KEY
API key for ElevenLabs voice services, from the ElevenLabs dashboard.
DEFAULT_MIMO_API_KEY
Other optional settings (40)
BYOK_KEY_ID
Key ID paired with the BYOK private key.
BYOK_ALLOW_EPHEMERAL_KEY
Allow ephemeral BYOK keys; true for local runs, false in production.
DEPLOYMENT_MODE
Deployment mode, local or hosted; hosted deployments should set hosted.
ALLOW_LOCAL_NETWORK_PROXY
RATE_LIMIT_STORE
DOCUMENT_PARSE_JOB_STORE
PLUGIN_REGISTRY_STORE
UPSTASH_REDIS_REST_URL
DEFAULT_PROVIDER_TYPE
Server default provider type: Google, Anthropic, OpenAI, or OpenAI Compatible.
DEFAULT_PROVIDER_NAME
DEFAULT_PROVIDER_BASE_URL
DEFAULT_PROVIDER_MODELS
DEFAULT_MODEL_TITLE_GENERATION
DEFAULT_MODEL_RELATED_QUESTIONS
DEFAULT_MODEL_CONTEXT_COMPRESSION
DEFAULT_MODEL_PROMPT_OPTIMIZATION
DEFAULT_MODEL_RAG_QUERY
DEFAULT_MODEL_MEMORY
DEFAULT_SEARCH_PROVIDER
DEFAULT_SEARCH_BASE_URL
DEFAULT_RAG_BASE_URL
DEFAULT_RAG_TOP_K
DEFAULT_RAG_CHUNK_SIZE
DEFAULT_RAG_NAMESPACE
DEFAULT_DOCUMENT_PARSE_PROVIDER
DEFAULT_VOICE_PROVIDER
DEFAULT_ELEVENLABS_STT_MODEL
DEFAULT_ELEVENLABS_TTS_MODEL
DEFAULT_ELEVENLABS_TTS_VOICE_ID
DEFAULT_MIMO_STT_MODEL
DEFAULT_MIMO_TTS_MODEL
DEFAULT_MIMO_TTS_VOICE_ID
DEFAULT_SYSTEM_PROMPT
DEFAULT_ENABLE_AUTO_TITLE
DEFAULT_ENABLE_RELATED_QUESTIONS
DEFAULT_ENABLE_AUTO_COMPRESSION
DEFAULT_ENABLE_CODE_COLLAPSE
DEFAULT_ENABLE_HTML_VISUAL_PROMPT
NEXT_PUBLIC_SITE_URL
Public site URL used for metadata and generated public links.
TRUST_PROXY_HEADERS
Local run requires Node.js 22 and pnpm 10.30.3. Use pnpm install then pnpm dev, open http://localhost:3000, and configure at least one model provider in Settings. Deployment options include Docker Compose (docker compose up --build), Docker image (docker build -t neo-chat:local .), Vercel, and Cloudflare Workers.
Check that it works
Run pnpm install && pnpm dev, open http://localhost:3000, and configure at least one model provider in Settings; a successful chat reply confirms the install. For MCP, install a remote MCP server from the plugin market and confirm its tools appear in the tool list.
Troubleshooting
- Check Node.js and pnpm versions meet requirements (Node 22, pnpm 10.30.3)
- Ensure at least one model provider is configured
- Check browser console and server logs for errors
- Inspect /api/health endpoint for deployment health status
- Verify environment variables like BYOK_ALLOW_EPHEMERAL_KEY are set correctly
- For Docker, ensure port mapping and network configuration are correct
Things to try
Once connected, you can ask your AI assistant things like:
- Install a remote MCP server from the registry and list its tools
- Start a new chat with an OpenAI-compatible provider and generate an image
- Upload a PDF to the knowledge base and ask questions about it
- Search my recent chats for notes about the project plan
Use cases
Local-first AI chat with data ownership
Multi-provider chat (Google, Anthropic, OpenAI, and compatible endpoints)
Extending functionality with skills, plugins, and MCP servers
Knowledge-base RAG and local global search
End-to-end encrypted cross-device sync
Offline PWA support
Overview
Neo Chat is a self-hostable, local-first AI chat application built with Next.js, React, TypeScript, and Zustand. It brings multi-provider chat, assistant presets, text-only Skills, OpenAPI-style plugin tools, remote MCP servers and an optional local stdio bridge, web and local global search, knowledge-base RAG, opt-in encrypted cross-device sync, versioned backup and restore, local memory, an offline PWA, voice, generated media, rich message rendering, citations, and editable artifacts into one clean workspace. It is designed for people who want the power of modern AI workspaces without giving up local data ownership. Chat history, workspace metadata, skills, plugin configuration, memories, search indexes, and files stay in the browser by default; server routes act as controlled proxies for model providers, web search, RAG, document parsing, voice, plugin and MCP execution, and deployment health.
Similar servers
Hevy MCP
77 · B
Manage your Hevy workout data from AI assistants via MCP.
Firecrawl MCP Server
75 · B
Firecrawl's official MCP server for web search, scraping, and structured extraction for AI agents
BioMCP
75 · B
One binary. One grammar. Evidence from the biomedical sources you already trust.
Compartment
75 · B
Encrypted, fully offline long-term memory for AI agents
Source revision 2c03fe88a57a Data synced 2026-10-19 Read the FMRS scoring method