You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
AI <-->|"2. LLM in the cloud communicates with a browser AI agent to act on web content"| BA
42
42
BA <-->|"3. Browser agent uses WebMCP tools to actuate the current page"| WMCP
43
43
WMCP -->|"4. WebMCP tools update UI and make API calls"| TP
44
44
```
45
45
46
46
#### Direct backend MCP flow
47
47
48
-

48
+
```mermaid
49
+
graph TD
50
+
AI["<b><i>AI agent platform</i></b>"]
51
+
52
+
subgraph WB["<b><i>Web Browser</i></b>"]
53
+
BIA["Browser-integrated AI agent"]
54
+
RP["Running Page <index.html>"]
55
+
end
56
+
57
+
subgraph TP["<b><i>Third-party service (example.com)</i></b>"]
58
+
MCP[("MCP Server")]
59
+
end
60
+
61
+
RP <-->|1. Browser loads page over HTTP| TP
62
+
BIA -->|2. User prompt sent to agent platform in the cloud.| AI
63
+
AI -->|"3. Agent platform uses pre-configured MCP server to interact directly with service and fulfill user request."| MCP
64
+
MCP -->|4a. MCP response routed back to agent platform.| AI
65
+
AI -->|5. Response rendered to user by browser agent. Web page has no direct visibility or control.| BIA
66
+
RP <-.->|4b. Service manually pushes updates to page.| TP
67
+
```
49
68
50
69
Many challenges faced by assistive technology also apply to AI agents that struggle to navigate existing human-first interfaces when agent-first "tools" are not available. Even when agents succeed, simple operations often require multiple steps and can be slow or unreliable.
0 commit comments