auxilia

auxilia is an open-source web MCP client for teams. It lets you host AI assistants backed by remote Model Context Protocol  servers, share them across your organization, and keep every integration — tools, credentials, observability — under your own infrastructure.

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 — workspace admins register MCP servers once; every teammate can bind them to agents
• Remote MCP only — designed for server-to-server MCP connections over Streamable HTTP, not local stdio processes
• Self-hosted — runs on your own infrastructure, keeps credentials and conversation data in your database
• MCP-native — everything the LLM sees (tools, skills, resources) flows through MCP, so the core stays small

What you can build

• A CRM agent that queries HubSpot and drafts Slack replies
• A data analyst that runs BigQuery, produces charts in a sandbox, and posts results to a thread
• An on-call assistant that reads Sentry issues, searches Linear, and opens pull requests via GitHub
• A workspace-wide coordinator agent that dispatches work to specialized subagents

Architecture

auxilia is composed of three services plus an optional sandbox runtime:

┌─────────────┐     ┌─────────────┐     ┌──────────────────┐
│   Browser   │────▶│   Next.js   │────▶│     FastAPI      │
│   (React)   │◀────│   (proxy)   │◀────│   (LangGraph)    │
└─────────────┘     └─────────────┘     └────────┬─────────┘
                                                 │
                                    ┌────────────┼────────────┐
                                    │            │            │
                         ┌──────────▼──┐ ┌───────▼──────┐ ┌───▼──────────┐
                         │ Remote MCP  │ │  PostgreSQL  │ │  OpenSandbox │
                         │  Servers    │ │  + Redis     │ │  (optional)  │
                         └─────────────┘ └──────────────┘ └──────────────┘

Key concepts

Agents

An agent is defined by a system prompt, an avatar (emoji + color), a set of MCP server bindings, and optional subagents. Each agent can use tools from multiple MCP servers at once.

MCP servers

auxilia connects to remote MCP servers over HTTP (Streamable HTTP transport). Servers are registered at the workspace level by admins and can be bound to any agent.

Tools

Each tool exposed by an MCP server can be set to one of three states per agent: always allow, needs approval, or disabled. Approvals happen inline in the chat (or in Slack).

Sandbox

Turn on the sandbox for an agent to give it a Linux code-execution environment with ls, read_file, write_file, edit_file, glob, grep, and execute. Useful for data analysis, scripting, and file manipulation tasks.

Workspace roles

• admin — manage MCP servers, invites, personal access tokens, agent permissions
• editor — create and edit agents
• member — chat with agents the admin has shared with them

Supported LLM providers

Configure one or more providers via environment variables. At least one is required.

Integrations

• Slack — invoke workspace agents, approve tool calls, and stream responses from Slack threads
• Langfuse — trace LLM calls, tool calls, and costs per agent and user

Next steps

• Get Started — run auxilia locally with Docker Compose
• Deployment — deploy on Google Cloud Run or any Docker host
• MCP Servers — register workspace MCP servers
• Agents — create and configure agents
• Sandbox — enable code execution for an agent