convez.ai
MCP Server
keurcien

auxilia

keurcien/auxilia

Ver no GitHub
BomPluginScore · de 0 a 100

A lightweight web MCP client designed to host your MCP-powered AI assistants. auxilia is designed to only support remote MCP servers.

Instalação

$npx convez-ai keurcien/auxilia

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
24
Forks
5
Acompanhando
0
Saúde
42%
Último commit
há 3 dias
Versão
backend-v0.9.27
Issues abertas
49
Contribuidores
7

README

A tradução para o português deste README ainda está na fila.


https://github.com/user-attachments/assets/3236f9da-28c5-44c8-8c2c-4c68199f187c

Why auxilia

Most MCP clients today are desktop apps tied to a single user. auxilia is different:

  • 🌐 Web-based — accessible from any browser, no desktop app to install
  • 👥 Multi-user — admins register MCP servers once; every teammate can bind them to agents
  • 🔌 Remote MCP only — built for server-to-server Streamable HTTP connections, not local stdio
  • 🏠 Self-hosted — runs on your infrastructure; credentials and conversation data stay in your database
  • 🧩 MCP-native — tools, skills, and resources all flow through MCP, keeping the core tiny

Learn more about auxilia: https://auxilia-docs.vercel.app/

✨ Features

🤖 Agents

  • Define agents with a system prompt, avatar (emoji + color) and bound MCP servers
  • Subagents — coordinator agents can dispatch work to specialized subagents
  • Workspace roles & per-agent permissions — owner, admin, editor, user, plus private/shared agents
  • Streaming responses over SSE via the LangGraph SDK, with full LangGraph checkpointing

🔧 Tools & MCP Servers

  • Register remote MCP servers at the workspace level — everyone can reuse them
  • Per-tool control: always allow, needs approval, or disabled
  • Human-in-the-loop approvals inline in chat or Slack
  • AES-GCM encryption for API keys at rest, per-user OAuth 2.1 token storage in Redis
  • Built-in connectivity probes with automatic token refresh

🧪 Code Sandbox

Turn any agent into a lightweight data-analyst with an isolated Linux environment powered by OpenSandbox:

  • Filesystem tools: ls, read_file, write_file, edit_file, glob, grep
  • Shell access via execute
  • Lazy-spawned containers with a 30-minute TTL, reconnectable across browser refreshes

⚠️ To enable code execution, you first need to deploy Opensandbox (either on a simple VM or Kubernetes), and set it up in auxilia.

💬 Slack Integration

  • Start a chat with auxilia and pick an agent
  • Tool-approval buttons rendered with Block Kit
  • Slack ↔ auxilia users matched by email

📊 Observability

  • Langfuse integration — trace every LLM call, tool call, and cost per agent and user
  • Request-scoped profiling via RequestTimer (see app/utils/timer.py)

🔐 Authentication

  • JWT sessions over HttpOnly cookies
  • Google OAuth sign-in (Authlib)
  • Personal Access Tokens (PATs) for programmatic access
  • Argon2 password hashing

🚀 Quick Start

Prerequisites

  • Docker and Docker Compose
  • At least one LLM API key (Anthropic, OpenAI, Google, or DeepSeek)

Run with Docker

git clone https://github.com/keurcien/auxilia.git
cd auxilia
cp .env.example .env   # add your LLM API keys
make build && make up

Open http://localhost:3000. The first account you create becomes the workspace admin.

Run for development

cp .env.example .env
make dev

make dev starts PostgreSQL, Redis, the FastAPI backend (with migrations applied) and the Next.js frontend in parallel — all with hot reload.

See the Get Started guide for the full walkthrough.

🤝 Integrations

LLM Providers

ProviderModels
AnthropicClaude Haiku 4.5, Sonnet 4.6, Opus 4.6
OpenAIGPT-4o-mini
GoogleGemini 3 Flash Preview, Gemini 3 Pro Preview
DeepSeekDeepSeek Chat, DeepSeek Reasoner

MCP Servers — one-click install

Notion · Linear · GitHub · HubSpot · BigQuery · Slack · Sentry

Plus any custom remote MCP server — just paste a URL and your OAuth credentials or API keys.

Workspace integrations

  • Slack — invoke agents, approve tools, stream replies in threads
  • Langfuse — LLM + tool-call tracing and cost attribution
  • Google OAuth — SSO for your workspace

💡 What you can build

  • A CRM agent that queries HubSpot and drafts Slack replies
  • A data analyst that runs BigQuery and stores the analysis in a Notion document
  • An on-call assistant that reads Sentry issues, searches Linear, and opens PRs via GitHub
  • A workspace coordinator that dispatches work to specialized subagents

📚 Documentation

🗺 Roadmap

  • Scheduled tasks
  • Skills
  • Support for more sandboxes (e.g. Daytona)
  • Deployment guides
  • More SSO providers (Okta, Entra ID)

Have an idea? Open an issue or start a discussion.

🤝 Contributing

Contributions are very welcome! Whether it's a bug report, an agent idea, a new MCP server recipe, or a UI improvement:

  1. Fork the repo and create a branch (git checkout -b feature/my-change)
  2. Run make dev and make sure tests pass (cd backend && uv run pytest)
  3. Open a pull request

Please read CLAUDE.md in the root of the repo — it documents the backend's layered architecture (router → service → repository → model) and naming conventions.

📄 License

AGPL-3.0 — free to use, modify, and self-host.


Ver o original no GitHub

Bom para

Relacionados