airfocus Adds Bidirectional MCP to Product Platform

On June 1, 2026, airfocus, the product management tool acquired by Lucid two years ago, announced via Yahoo Finance the launch of its Product Intelligence Platform, a major update that includes native bidirectional MCP support and what the company calls a Connected Intelligence Layer. This is no minor announcement: it is the first vertical product management platform to openly commit to MCP as the axis for integration with external agents, rather than building a closed proprietary system.

Until now, AI integration in product tools like Jira, Linear, or airfocus itself followed a unidirectional pattern: the LLM read system data, generated text, and users manually pasted it back. airfocus proposes a different approach: through an MCP server exposed by the platform, an agent—whether Claude Code, a custom subagent, or any compatible client—can read roadmap status, write updates, create items, or modify priorities directly, without human intervention at each step.

What bidirectional MCP means in this context

The Model Context Protocol, a standard championed by Anthropic, defines how an LLM calls external tools through a structured interface. Most current implementations are read-only or single-action: the agent queries data or executes a specific function. What airfocus calls bidirectional means the MCP server can also emit notifications or updates to the agent proactively, for instance alerting when a feature changes status while the agent works on a related task.

This aligns with a pattern we've seen growing across the Claude ecosystem for months: MCP servers that not only respond to calls but maintain a persistent context channel with the agent. It is the same principle some experimental real-time database MCP servers use, now applied to a product tool for everyday use.

The Connected Intelligence Layer and who benefits

The other piece of the announcement is the connected intelligence layer, which airfocus describes as a set of internal models trained on product data—user feedback, usage metrics, prioritization patterns—that agents can query via MCP to get enriched context, not just raw data.

This is particularly relevant for product teams already using Claude Code or custom agents for analysis or spec-writing tasks. With this MCP available, a subagent could handle the complete cycle: read accumulated feedback for a feature, consult airfocus's intelligence layer, propose a priority update, and write it to the roadmap, all within a Claude Code session without the PM opening the airfocus interface.

For small teams or startups, the value is clear. For larger organizations with stricter validation processes, the obvious risk cuts the other way: a misconfigured agent could modify the roadmap without sufficient oversight. This is where Claude Code hooks, specifically `PreToolUse`, can play a control role: intercept the MCP call before it executes a write operation and request confirmation or logging.

Ecosystem context

This move from airfocus comes as the catalog of third-party MCP servers grows at a steady pace. The difference from earlier integrations is that we are not looking at a basic read-only connector but an MCP server designed from the outset to support agents with real autonomy. That a product software company, not a technical infrastructure vendor, is making this bet suggests the model of an agent acting on business tools is shifting from an engineering experiment to a product requirement.

Editor's note: airfocus's technical approach is sound and the timing makes sense. What remains to be seen in the coming weeks is whether the MCP server implementation is robust enough to withstand real-world use in agent pipelines, or whether it stays as a demo feature. The bidirectionality, in particular, promises more than most current MCP servers deliver.