Obot

Obot is an open-source, self-hosted platform for the Model Context Protocol. It provides MCP hosting, MCP registries, an MCP gateway, and a chat platform in a single system.

The Problem

MCP provides a standard way to connect AI applications to tools, services, and data. Running MCP in practice introduces a set of common problems:

• Discovery: Users need a clear way to find available MCP servers without relying on ad hoc sharing.
• Security: Servers need to be authenticated, access needs to be controlled, and activity needs to be auditable.
• Operations: Servers must be deployed, updated, and scaled without manual coordination.
• Policy Enforcement: Requests sometimes need to be inspected or blocked before they reach downstream systems.

Obot addresses these problems by providing a complete, self-hosted MCP platform.

Getting Started

To run Obot locally, start it with Docker:

docker run -d --name obot -p 8080:8080 \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -e OPENAI_API_KEY=<API KEY> \
  ghcr.io/obot-platform/obot:latest

Open http://localhost:8080 in your browser to access the Obot UI.

Note: Replace <API KEY> with your OpenAI API key. You can also set ANTHROPIC_API_KEY or configure model providers through the admin UI.

For additional installation options, see the Installation Guide at https://docs.obot.ai/installation/general.

Platform Components

Obot is built around four core components.

MCP Hosting

Run and manage MCP servers directly within Obot:

• Run MCP servers locally with Docker or deploy them to Kubernetes
• Support for Node.js, Python, and container-based servers
• Support for both single-user STDIO servers and multi-user HTTP servers
• Controls for who can deploy servers, publish them to the catalog, or share them
• Built-in OAuth 2.1 and token handling for authentication

MCP Registry

A central place to list and discover MCP servers:

• Curated catalog of available MCP servers
• Shared credentials and authentication handled by the platform
• Conformance with the MCP registry specification
• Server visibility based on user access

MCP Gateway

A single entry point for accessing MCP servers:

• Access rules for users and groups
• Logging of MCP requests and responses
• Usage visibility to understand which servers are being used
• Request inspection and filtering before requests reach servers

Obot Chat

A chat client built to work directly with MCP:

• Support for multiple model providers including OpenAI and Anthropic
• Add domain-specific information to conversations with built-in RAG
• Project-wide memory to maintain important context across conversations for personalized interactions
• Create and share reusable project configurations with other users
• Scheduled tasks for recurring workflow automations

How the Pieces Fit Together

1. Platform owners or administrators can manage MCP servers in the registry and define access rules.
2. MCP servers are deployed and run using the hosting layer.
3. All MCP traffic flows through the gateway, where access control, policy checks, and auditing is applied.
4. Users interact with MCP servers through Obot Chat or other MCP-compatible clients.

Technical Overview

• Self-Hosted: Deploy on your own infrastructure for complete control over data and security
• MCP Standard: Built on the open Model Context Protocol for maximum interoperability
• Security-First Design: OAuth 2.1, encryption at rest and in transit, comprehensive audit logging
• Extensible: Easy integration with custom tools, services, and existing systems
• GitOps Ready: Manage catalog and configuration as code

Documentation

Documentation is available at https://docs.obot.ai.

Community

• Documentation: https://docs.obot.ai
• Discord: https://discord.com/invite/9sSf4UyAMC

License

Obot is open-source software. See the LICENSE file for details.