As of 2026-09-24, 196 of the 330 tools in the catalogue document an integration with AI agents. The agent integration fact names an MCP server or client for 190 tools. Among categories, MCP servers and agentic end-to-end testing account for the most documented integrations.
What the agent integration fact records
The agent integration fact describes how a tool connects to AI agents and coding assistants. Examples are an MCP server, an IDE plugin, or a documented connection to an agent such as Claude Code, Cursor or Copilot. The value is a short description taken from the source page. Like every fact, it carries an evidence level, a source and a verified-on date.
Evidence levels
Counts across 330 tools as of 2026-09-24:
| Evidence | Tools |
|---|---|
| Documented | 196 |
| Not documented | 99 |
| Not recorded | 26 |
| Vendor-stated | 9 |
Not recorded means the tool file has no agent integration fact. Not documented means we looked and found no integration on public pages.
MCP mentions
The agent integration fact mentions MCP (Model Context Protocol) for 190 tools as of 2026-09-24. Of these, 181 are documented and 9 are vendor-stated. All 9 vendor-stated integrations mention MCP.
Of the 196 documented integrations, 15 describe a connection without mentioning MCP.
The MCP mentions take two forms in the data. Some tools are themselves MCP servers. For example, a11y MCP is an MCP server that exposes axe-core accessibility checks to MCP clients, per its repository. Other tools offer an MCP server next to their own platform. For example, accessFlow documents an MCP server that connects AI clients to its audit results, per its support article.
Documented integrations by category
The table counts tools with a documented agent integration, by primary category. Counts are as of 2026-09-24. Rows are sorted by the documented count, then by category name A-Z. Shares are rounded to whole numbers.
| Category | Documented | Tools in category | Share |
|---|---|---|---|
| Agentic end-to-end testing | 35 | 54 | 65% |
| MCP servers for testing | 28 | 28 | 100% |
| Mobile testing with AI | 21 | 30 | 70% |
| Natural-language test authoring | 17 | 33 | 52% |
| AI test management | 15 | 29 | 52% |
| API testing with AI | 12 | 21 | 57% |
| Autonomous exploration | 10 | 20 | 50% |
| IDE and repo test agents | 10 | 14 | 71% |
| Results analysis and triage | 10 | 13 | 77% |
| Accessibility testing with AI | 9 | 11 | 82% |
| Unit test generation | 6 | 17 | 35% |
| Performance testing with AI | 5 | 8 | 63% |
| Test data generation | 4 | 11 | 36% |
| Test maintenance and self-healing | 4 | 11 | 36% |
| Visual AI testing | 4 | 6 | 67% |
| Session replay testing | 3 | 3 | 100% |
| Monitoring with AI checks | 2 | 4 | 50% |
| Testing assistants | 1 | 6 | 17% |
| Managed AI QA services | 0 | 11 | 0% |
The counts sum to 196.
What the category figures show
Agentic end-to-end testing has the highest count, with 35 tools. It is also the largest category in the catalogue, with 54 tools.
Every tool in the MCP servers category documents an agent integration. This follows from the category definition: MCP servers exposing browsers, test runners, bug trackers or test management to AI agents.
Session replay testing also documents an integration for all of its tools, but the category has only 3 tools.
None of the 11 managed AI QA services documents an agent integration as of 2026-09-24.
The counts use the primary category only. A tool with a secondary category counts once, under its primary category.
Pages that filter on this fact
The MCP integration feature page lists tools whose agent integration fact names an MCP server or client.
The AI coding agent workflow page lists tools whose agent integration fact names an MCP server, an IDE plugin or a coding agent. The named agents are Cursor, Claude Code, Copilot, Windsurf, VS Code and JetBrains.
Both pages list documented matches in the main table. Vendor-stated matches are listed separately, with their source and date.
Sources
- a11y MCP repository
- How to connect AI clients to accessFlow with MCP
- Tool pages: a11y MCP, accessFlow
- Counts come from the catalogue data behind All tools and the category hubs.