#1 Agent access security brokers (AASB)17/18 · Last reviewed

Unbound

Unbound is an agent access security broker for AI coding agents. It finds every agent, MCP server and skill on the company's machines, and decides each terminal command, file access and MCP call against the security team's policy before it runs.

Verdict: The strongest fit for governing coding agents today: fleet-wide visibility and point-of-action policy for terminal and MCP, with an audit trail, without taking developers' agents away.

What it covers

  • Coding agents. Claude Code, Cursor, Codex, GitHub Copilot, Gemini CLI, Windsurf, Cline and more than 20 others on corporate-managed devices, plus the Claude Desktop app for people outside engineering.
  • Terminal commands. Each command an agent runs is judged against central policy and can be audited, warned on, blocked or held for a person's approval, in Slack where set up. The developer sees why, in plain words, in the terminal.
  • MCP servers. Discovers every MCP server connected to every agent, flags unofficial or risky ones, and sets policy per team, per server or per tool call.
  • Data sent to model providers. Policy on sensitive data in what agents send to model providers, observed at the workstation.
  • Audit evidence. A record of every decision, with search and export; SIEM export to tools such as Splunk and Datadog on the Enterprise plan.

What it does not do

  • Not a code scanner. It does not run SAST, SCA or secrets scanning on the code agents write. Keep your AppSec scanning in CI.
  • Not IAM. It does not issue, scope or rotate credentials. Least privilege and short-lived tokens stay with your identity provider and secrets manager.
  • Not EDR. It does not detect malware or attacker techniques on the endpoint. Keep your EDR.
  • Not an AI gateway. Model traffic does not go through Unbound; requests go straight to the model provider. It works alongside gateways such as LiteLLM and Portkey.
  • Not a reader of your source code. It monitors agent behaviour metadata, such as which tool was invoked, not the content of your source code.

Scored on the same criteria

The same six criteria, and the same 0 to 3 scale, as every approach in the ranked guide.

CriterionScoreWhy
Visibility 3Inventories every agent, IDE plugin, MCP server and skill on the fleet, with who installed it and what it can reach.
Enforcement 3Audit, warn, block or require approval for each command, file access and MCP call, from central policy.
Developer friction 3Developers keep their agents; warnings appear in the terminal only for risky actions, with a reason and a safer alternative.
Coverage of terminal and MCP actions 3Terminal commands and MCP calls across more than 20 agents.
Audit evidence 3Decision log with search and export; 30-day retention on Pro, 90 days or more with SIEM export on Enterprise.
Deployment effort 2One MDM configuration to roll out; the security team still chooses and tunes the policy.
Total17/18Listed under Agent access security brokers (AASB), ranked #1.

Who it is for

Security and platform teams at companies where developers use Claude Code, Cursor, Codex, Copilot or other coding agents on company-managed machines, and who need to allow them with control rather than ban them.

How it is deployed

  • Uses the hooks and managed settings coding agents already ship; developers do not install a new agent.
  • Rolled out with a short MDM configuration for Jamf, Intune or Kandji.
  • Policies follow existing groups from identity providers such as Okta and Microsoft Entra.

Price

  • Business Pro: $18 per user a month, no user minimum.
  • Enterprise: from $28 per user a month, with SSO, SCIM, approval workflows and SIEM export.
  • Community edition, for one developer's own agents, has a free tier.
  • A free two-week assessment for teams with 50 or more developers using coding agents.

As listed on its pricing page on 16 November 2026. Check the current pricing before you budget.

Sources