The six agents we track
Our census tracks six agents: Claude Code, Cursor, GitHub Copilot, OpenAI Codex, Windsurf and Claude Desktop. They are the agents vendors in this category name most often on their own pages. For each vendor we record whether the cited page names the agent. We do not record an agent that is only implied by a general phrase such as coding assistants.
What the census found
On the pages we read, Zenity names four of the six: Claude Code, Cursor, GitHub Copilot and Codex, and also names Cowork. Bay names Claude Code, Codex and Claude Desktop, for which it describes managed settings. Noma Security names Claude Code, Cursor and Codex. Prompt Security names GitHub Copilot, Cursor and Claude Code. Lasso Security names Claude Code for its hooks integration and discovers MCP servers across Cursor and Windsurf. A Palo Alto Networks blog about Koi names Claude Code, Cursor and Codex. Harmonic Security's pricing page names Claude Code and GitHub Copilot among the developer tools its Guide tier covers. Onyx Security and Bloom Security describe agents in general terms without naming them on the pages we read.
What a name proves, and what it does not
A name on a page proves the vendor wants to be considered for that agent. It does not tell you whether the product inventories the agent, sets its managed settings, or decides on its tool calls before they run. Those are different depths of support, and they can differ between agents in the same product. Bay's own wording, for example, limits in-line enforcement to supported agent surfaces.
The reverse also holds. A vendor that names no agents may still control them; it has simply not said so publicly. That is why named agent coverage carries only 5% of our score.
Three depths to ask about
- Seen. The product finds the agent on devices, with its version, its MCP servers and its settings.
- Governed. The product can set or lock the agent's configuration: permission rules, allowed MCP servers, plugin sources.
- Decided. The product sees each tool call or command from the agent before it runs and can return a verdict.
Ask the vendor to fill in these three columns for every agent your people use.
Agents change faster than pages
New agents and new versions arrive every month, and hook interfaces change with them. A support list from the spring may be out of date by the autumn. Ask how quickly the vendor added support for the last major agent release, and whether support for a new agent version needs a product update on the device.
Using the estimator
The agent risk estimator lets you tick the agents you run and filter to vendors that name all of them. Use it to shortlist, then run the three-depth check with each shortlisted vendor.