> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ultra.security/llms.txt
> Use this file to discover all available pages before exploring further.
# Introduction
> Ultra is a security proxy for the Model Context Protocol (MCP)
Ultra sits between your MCP clients and upstream MCP servers, providing complete visibility into every tool call, resource read, and prompt request flowing through your AI development stack.
A security-first proxy that records every action for compliance and audit, then enforces policy inline with guardrails and governance rules on every tool call.
Use any AI client with any MCP server. Ultra is the universal bridge between your agents and tools. Manage all your connections from one place. Ultra probes and normalizes MCP servers automatically, fixing quirky startup flags and protocol issues so you don't have to debug them yourself.
Full traces, audit logs, and metrics for every MCP operation. Know exactly what your AI agents are doing.
Centralized management for teams. Register gateways, sync telemetry, and manage workspaces from one control plane.
## Local-First Architecture
Ultra runs entirely on your local machine as a lightweight binary. There are no remote proxies routing your MCP traffic through external servers. Your MCP clients launch Ultra as a local subprocess, and all processing happens on-device.
This means:
* **No latency overhead** -- Ultra runs as a local process, not a remote service. MCP clients launch it as a subprocess on your machine.
* **Works offline** -- All observability (traces, audit logs, metrics) is stored locally on your machine. No internet connection required.
* **Your data stays yours** -- Tool calls, responses, and audit trails live on your machine at `~/.config/ultra/ultra.db`. Nothing is sent externally unless you enable Hub sync or OTLP export.
## Why Ultra
**MCP connections are unmanageable at scale.** Every AI client needs its own connection to every MCP server. Five clients and ten servers means fifty configurations to maintain, update, and secure. Ultra is the universal bridge. Connect each client to Ultra once, and Ultra handles every server behind it. Add a server in one place and every client gets access immediately.
**AI agents are powerful, but opaque.** When Claude or Codex calls an MCP tool, you have no visibility into what's happening. Ultra changes that by sitting in the middle and recording everything (traces, audit events, and metrics) so you always know what your AI agents are doing.
**Teams need centralized visibility.** As organizations adopt MCP across multiple developers, managing dozens of MCP server configurations and monitoring tool usage becomes unmanageable. Ultra Hub provides the control plane: register gateways, sync telemetry, and manage it all from one dashboard.
**Security starts with observability.** You can't secure what you can't see. Ultra gives you the complete audit trail, then builds enforcement on top: [guardrails](/hub/guardrails) and [governance rules](/hub/governance) evaluate every tool call inline and enforce policy before a request reaches an upstream server.
## What You Get
### Ultra Server
A transparent MCP proxy that connects multiple agents to multiple upstream servers and presents them as one. Supports stdio, HTTP/SSE, and Streamable HTTP transports. Hot-loads new servers without restarting.
### Dashboard
View traffic, trace details, audit logs, and server status in real time. [Ultra Hub](/hub/overview) provides a centralized dashboard for teams to monitor activity across all gateways from one place. CLI users can also access a local dashboard at `http://localhost:1566`.
### Observability
Every MCP operation is recorded automatically:
* **Traces**: Full request/response payloads, timing, status, and OpenTelemetry span IDs
* **Audit events**: Security-relevant events with severity, principal identity, and outcome
* **Metrics**: Request counts, latency histograms, error rates, and per-server breakdowns
Data is stored locally by default, with optional export to any OpenTelemetry-compatible backend (Datadog, Grafana, etc.).
### Anomaly Detection
AI-powered analysis that scores every tool call for suspicious behavior including unusual access patterns, sensitive data exposure, prompt injection, and data exfiltration. Configurable alert thresholds and risk categories.
### Guardrails & Governance
Inline enforcement on every tool call. [Guardrails](/hub/guardrails) evaluate requests against built-in and custom rules to protect against risks like credential leakage, sensitive data exposure, and cross-server exfiltration. [Governance](/hub/governance) rules determine which tool calls are allowed. Both are configured in Ultra Hub and synced to every gateway.
### Built-in MCP Tools
Ultra exposes its own tools that your AI client can call directly. [Search for servers, add them, migrate existing configs, and check status](/getting-started/ultra-mcp-tools) without leaving your conversation.
### Ultra Hub
Optional cloud control plane for teams. Register gateways, sync telemetry from across your organization, manage workspaces, and enforce role-based access control. Includes SSO and SCIM for enterprise identity management.
### Gateways
Each developer runs their own Ultra instance (gateway) that connects to Hub. Telemetry syncs automatically, with offline support. Data syncs when connectivity is restored.
### CLI
A full command-line tool for managing your installation. Search for servers, add them, run diagnostics, open the dashboard, and more.
```bash theme={null}
ultra start # Start the proxy
ultra search github # Find MCP servers
ultra add notion # Add a server
ultra servers # List connected servers
ultra dashboard # Open the web dashboard
ultra doctor # Run diagnostics
```
Get Ultra running in seconds
Understand the architecture