MCP & Browser Integration
Dynamic Chrome MCP port injection, browser session sharing, and MCP server management.
MCP Overview
MCP (Model Context Protocol) servers extend what AI agents can do by providing tools like browser automation, database access, and external API integration. AgentsInFlow manages MCP servers as part of the engine configuration, injecting them at execution time.
- MCP servers are configured per project in the engine settings
- Servers start automatically when an execution begins
- Currently supported for Claude Code engine
Chrome Session Sharing
AgentsInFlow dynamically injects the Chrome DevTools debug port into MCP server configurations. This allows multiple agents to share a single Chrome browser session for web automation tasks, without hardcoding port numbers.
The debug port is detected at runtime and passed to MCP servers as an environment variable. No manual port configuration is needed.
Runtime MCP Injection
MCP servers are injected into the engine configuration at execution time, not hardcoded into project files. This means:
- Different executions can use different MCP servers
- Server configurations stay in AgentsInFlow, not in your repo
- Dynamic values (ports, tokens) are resolved at launch time
Troubleshooting
Port Conflicts
If Chrome is already running with a debug port, AgentsInFlow detects the existing port. Close other Chrome debug sessions if you experience connection issues.
Session Errors
If a browser session becomes unresponsive, restart the Chrome instance from AgentsInFlow. The debug port will be re-detected on next execution.
MCP Server Not Found
Verify the server command and path in your project engine settings. MCP servers must be installed and accessible from your system PATH.