Ultra Security provides runtime security for AI agents. It gives security teams a control point between the agents their staff and engineers use, such as Claude Code, Cursor and GitHub Copilot, and the MCP connectors and tools those agents call, so every agent action can be seen, checked against policy and blocked when it crosses a line.
It covers both agents running on user devices and internal agents that teams run for the company, and is aimed at security teams who need to let staff use coding and productivity agents without losing track of what those agents touch.
At A Glance
| Vendor | Ultra Security |
|---|---|
| Category | Runtime security and governance for AI agents and MCP connectors |
| Identities covered | AI agents, the users, service accounts and devices behind them, connector credentials |
| Deployment | Single binary on each user’s machine, managed from the cloud-based Ultra Hub |
| Agentic AI | Agent discovery, MCP proxy, inline guardrails, anomaly detection |
| Customers | Parker, The Martec Group, Vantage, Michigan State |
| Certifications | ISO 27001:2022, SOC 2 Type 2, ISO/IEC 42001:2023 |
| Market classification | Agentic AI Identity |
What Ultra Security Does
As teams adopted AI agents, tool-connected clients became hard to audit. The Model Context Protocol widened what an agent can reach, and with it came risks that only show up at runtime: poisoned tool descriptions, prompt injection, credentials leaking into tool parameters and a sprawl of connections nobody approved. Ultra was built to watch agent behaviour and enforce controls while agents are working, putting enforceable boundaries around them without slowing down the work they do. Every action is attributed to a person or device, so accountability survives automation.
Products
Discover
Discover builds an inventory of the agents and connectors in use across the organisation, including those built into Claude and Cursor and connectors to services such as GitHub, Notion, Linear and HubSpot. When a new agent or connector appears, it falls under policy automatically, so shadow agents do not stay unmanaged.
Observe
Observe intercepts each agent action and ties it to the identity, device, connector and tool involved. Over time it builds a baseline of normal usage, and every event carries an audit record showing which rule allowed or blocked it, which gives security and compliance teams an answer to “what did this agent do, and why was it allowed?”
Protect
Protect has three parts:
| Govern | Defines what each agent may do, down to individual tools, before any action runs |
|---|---|
| Secure | Prebuilt guardrails against tool poisoning, prompt injection, credential exposure and rug pulls, plus custom rules |
| Detect | Raises alerts on anomalous activity such as unusual access or data movement, without blocking it |
Ultra Hub
Ultra Hub is the cloud management side of the platform, where administrators manage devices, teams, workspaces, guardrails and access. Enforcement is configured once in the Hub and synced to every device.
How It Works
Ultra sits between agents and the tools they call. It acts as a transparent proxy between MCP agents and upstream MCP connectors, presenting the combined tools, resources and prompts of all connectors as a single server, so agents connect once and reach every connector through it. Each action is recorded, checked against policy and routed to the right connector.
| Local binary | Ultra runs on each user’s machine as a single binary that agents launch as a subprocess; requests are processed on the device, and rollout can be by direct install or MDM |
|---|---|
| Interceptor pipeline | Every request passes through interceptors for tracing, logging, audit, metrics, governance and guardrails, and if a completed call cannot be written to the audit log, Ultra fails the call |
| Aggregator | Manages connections to multiple upstream connectors, namespaces tool names and routes calls over stdio, HTTP/SSE or Streamable HTTP, with OAuth2 for connectors that need it |
| Data storage | Observability data stays on the device by default, with optional background sync to Ultra Hub |
Identities And Systems Covered
| AI agents | 19 supported agents, including Amazon Q Developer, Claude Code, Claude Desktop, Cline, Codex, Cursor, Gemini CLI, GitHub Copilot, Goose, Roo Code, Windsurf and Zed |
|---|---|
| People, service accounts and devices | Members signed in through SSO, service accounts, devices authenticated by a deploy key and unauthenticated agents, each given an assurance level of Authenticated, Device-attributed or Unattributed |
| Credentials and secrets | Credential files, plus API keys, tokens and private keys appearing in tool parameters |
| Connectors | Upstream MCP connectors and the credentials agents use to reach them |
Agentic AI Capabilities
Built-In Guardrails
Guardrails and governance are enforced inline on every tool call. The built-in set covers credential and secret protection, PII and personal data protection, tool and server isolation, parameter validation and input sanitisation, in-line authorisation with destructive action blocking, and rate limiting with a circuit breaker. Each guardrail can block, monitor or redact; redaction masks the matched content in both the request and the response so the call can still go ahead.
Anomaly Detection
Anomaly detection runs as a background process in Ultra Hub that scans MCP traffic on a schedule, anywhere from every 15 minutes to every 24 hours, with on-demand scans as well. Each run covers the time since the previous one. Findings fall into categories such as enumeration, data exfiltration, injection and credential abuse, each rated from None to Critical.
Integrations
| Area | Integrations |
|---|---|
| Single sign-on | Okta, Microsoft Entra ID, Google Workspace, Auth0, JumpCloud, Microsoft AD FS, OneLogin, PingOne, Rippling, any OIDC or SAML 2.0 provider |
| Provisioning | SCIM 2.0, with provider groups synced to Ultra Hub teams |
| Observability and SIEM | Any OTLP backend, Amazon S3, Datadog, Grafana Tempo, Honeycomb, Jaeger, Panther |
| Device management | Jamf Pro |
| Alerting | Slack |
Company Background
| Founded | 2025 |
|---|---|
| Leadership | Chase Lee, Founder |
| Funding | A reported $10 million or more in seed funding at launch from stealth, from investors including Ludlow Ventures, Entrée Capital, Detroit Venture Partners, Firedrop, Hidden Capital and FDVC |
Note: This overview is compiled from Ultra Security’s publicly available information as of 3 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.