Minibridge

Minibridge is an MCP server proxy and bridge that securely connects AI agents to Model Context Protocol servers over the internet or locally. It allows developers, platform teams, and security engineers to wrap existing MCP tools inside a secure transport layer featuring TLS and client certificate validation. Minibridge introduces policy enforcement through policers, supporting local Open Policy Agent Rego policies or remote HTTPS endpoints for agent authentication, payload inspection, and content transformation. It also ensures tool integrity by checking Software Bill of Materials data and preventing unauthorized runtime mutations of tools or templates. In addition, Minibridge integrates observability into MCP workflows by generating Prometheus metrics and capturing OpenTelemetry traces, including span propagation via standard headers or tools like OpenInference. By operating in frontend, backend, or all-in-one modes, Minibridge enables organizations to expose and govern their MCP infrastructure safely without modifying downstream servers.

Category: AI & LLM Tooling

Tags: authentication, gateway, proxy, security, telemetry

Visit Minibridge

How to install and configure Minibridge

1. Install Minibridge using Homebrew with brew tap acuvity/tap && brew install minibridge, using AUR with yay -S minibridge, using Go with go install go.acuvity.ai/minibridge@latest, or by downloading a release binary from the GitHub repository. 2. Open your client configuration file, such as claude_desktop_config.json for Claude Desktop. 3. Wrap your existing MCP server command with Minibridge in all-in-one mode under mcpServers. For example: json { "mcpServers": { "fetch": { "command": "minibridge", "args": ["aio", "--", "uvx", "mcp-server-fetch"] } } } 4. If your client reports an execution error such as ENOENT, specify the absolute binary path obtained from which minibridge in the command attribute.

What you can do with Minibridge

• Securing internet-exposed MCP servers by enforcing TLS transport encryption and optional client certificate verification. * Intercepting MCP requests and responses with Open Policy Agent Rego rules to authenticate agents and sanitize tool inputs. * Validating runtime server integrity and dependencies using automated Software Bill of Materials verification checks. * Exporting tracing and telemetry data to OpenTelemetry collectors to observe agent interactions and span lifecycles across MCP calls. * Collecting real-time Prometheus monitoring metrics covering MCP server performance, request counts, and execution status.

Key facts

• https://github.com/acuvity/minibridge
• AI & LLM Tooling, Monitoring & Observability, Security & Compliance
• authentication, gateway, proxy, security, telemetry

Part of MCP Servers

What is Minibridge?

Minibridge is a security and observability proxy designed for Model Context Protocol servers. It operates as a bridge between AI agent clients and backend MCP servers, offering TLS termination, policy enforcement, telemetry reporting, and runtime integrity verification.

How do I install Minibridge?

You can install Minibridge on macOS using Homebrew via brew tap acuvity/tap and brew install minibridge. On Arch Linux, install it through the AUR with yay -S minibridge. You can also build it with the Go toolchain using go install go.acuvity.ai/minibridge@latest or download prebuilt binaries directly from GitHub releases.

How does Minibridge enforce security policies?

Minibridge integrates with policing services called Policers located in its backend architecture. Policers can evaluate inputs and outputs to authenticate agents or redact content, using either local Open Policy Agent Rego definitions or remote HTTPS webhooks.

Which MCP clients work with Minibridge?

Minibridge works with standard MCP clients that support launching external commands or connecting to MCP endpoints, such as Claude Desktop. In local setups, you simply replace the server execution command with minibridge aio followed by your target MCP tool.