From unstructured data to MCP servers

“We built our MCP server over a weekend and had 20x signups the first week. So, don’t just wrap APIs — design tools around outcomes.” — Kirk Marple

Most companies are sitting on mountains of unstructured data — emails, PDFs, transcripts, Slack threads, but struggle to make it usable for agents and apps. In this episode, Ravi Madabhushi talks with Kirk Marple, founder & CEO of Graphlit, about building developer-first infrastructure for turning messy content into structured, searchable, and agent-ready knowledge.

You’ll hear:

• Why metadata is the secret weapon for making unstructured content truly queryable.
• How customers use Graphlit for everything from search and RAG pipelines to derivative publishing like blog posts or audio summaries.
• Why MCP isn’t just about API wrappers — it’s about task-based tools and resource modeling.
• Practical lessons from building and scaling an MCP server, including auth headaches, cost traps, and early adoption learnings.
• Where MCP is headed: low-code/no-code apps, vertical MCP clients, and a new surface for developers and non-developers alike.

The conversation closes with advice for builders: design your MCP servers to work with the next generation of agent frameworks, avoid lock-in, and model your data consistently so applications can evolve on top of it.