Mint is an AI agent security company. Its platform helps security teams find the AI agents running across an organisation, watch what they do and stop risky actions while they run, covering coding and assistant agents such as Claude Code, Cowork and Copilot as well as agents built in house.
The problem Mint addresses is that agents, the MCP servers they call and the skills they load spread faster than security teams can track them. Each new agent brings new access, and few teams can say which agents hold it or what they did with it.
At A Glance
| Vendor | Mint |
|---|---|
| Category | AI agent security, discovery and runtime control |
| Identities covered | AI agents, MCP servers, agent skills and the users who run them |
| Deployment | Cloud platform |
| Agentic AI | Agent inventory, runtime policy, policy set up through chat |
| Certifications | SOC, ISO and GDPR |
| Market classification | Agentic AI Identity |
What Mint Does
Mint’s approach is to let agents run, watch them constantly, leave routine sessions alone and step in only when an action is risky. Every function works from one live view of agents, their connections and their interactions, and patterns that span sessions are used to anticipate what could go wrong next. Security teams end up with a record of what each agent did and a way to intervene, without becoming a blocker for the people who use agents to get work done.
Products
Discover
Discover builds the inventory: every agent, MCP server and skill in use, with its owner, connections, access and usage. It answers the first question a security team asks about agents, which is what is running and who is responsible for it.
Observe
Observe traces agent sessions and records every action, linking each one so it can be followed back to its source. When behaviour looks anomalous it raises an alert in real time, with the session evidence attached.
Control
Control applies policy while agents run. Toggles cover destructive operations, sensitive data and prompt injection, and the platform detects and stops agents that drift off script or try to tamper with their controls. Policies can be set up by chatting with Mint rather than writing rules by hand.
Adapt
Adapt turns findings into a daily to do list of risks, events to review and policies to update. Threats are ranked and mapped to the users and assets they affect, and many fixes can be applied in one click.
How It Works
Mint runs as a cloud platform that connects to the places agents work; a reported integration time is minutes.
| Inventory first | Discovery produces the list of agents, MCP servers and skills with their owners. |
|---|---|
| Shared view | Session tracing and runtime enforcement work from that same inventory, so every alert or blocked action is tied to a known agent and owner. |
| Feedback | Findings from monitoring and enforcement flow back into the daily plan in Adapt. |
Identities And Systems Covered
| AI agents | Coding and assistant agents including Claude Code, Cowork and Copilot, plus homegrown agents. |
|---|---|
| MCP servers | Included in the agent inventory and monitored with the agents that call them. |
| Skills | Agent skills are discovered alongside agents. |
| Users and assets | Threats are mapped to the people and assets they affect. |
Company Background
| Leadership | Or Sagy, co-founder and CEO; Tomer Shanny, co-founder and CTO. |
|---|---|
| Investors | Picture Capital and Accel. |
| Mission | “creating a simpler way to discover, monitor, and govern AI agents at scale” |
Note: This overview is compiled from Mint’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.