The Lore team has introduced the /share team architecture, which allows for converting Claude Code session transcripts into structured web pages accessible via direct links.
What Happened
Lore developers implemented a system that uses a local MCP server to access the file system and a hosting MCP server to manage the publishing process. The entire cycle involves an 11-step data upload to S3, followed by parsing and the creation of canonical URLs. To ensure reliable upload automation, a specialized Upload API Key (UAK) system was implemented instead of standard OAuth tokens.
Context
The solution relies on the Model Context Protocol (MCP), which allows for the integration of local tools and cloud storage into a unified agent workflow. This enables a shift from managing isolated files to managing context threads and skills.
Why It Matters for the Industry
The project demonstrates a fundamental shift in AI-native workflows: value is shifting from individual documents to reproducible agent execution flows. This lays the groundwork for a new category of software—Context Orchestration Platforms—where the objects of management are model states and skills rather than just files.
Why It Matters for Users
Developers gain an efficient tool for collaborative debugging and reviewing AI agent outputs. Local Claude Code sessions can now be easily transformed into readable artifacts, which is critical for team development and rapid analysis of agent performance via a web interface.
What Is Not Yet Known / Limitations
There are security and privacy risks associated with transferring data through external MCP servers and S3 cloud storage.
Sources
Author
Look at AI, Editorial Team