Vercel's New Coding Agent Takes Away Your MCP Tool List. It Sends the Same 15 Schemas.

Author(s): Chew Loong Nian – AI ENGINEER Originally published on Towards AI. fx declares seventeen built-in tools, and a full turn with a subagent host carries fifteen function schemas. I measured a stock four-server MCP setup at 37 tools and 22,226 bytes — the exact payload this design keeps out of the request. Every few weeks I go through the same small humiliation. I add an MCP server to a coding agent because it looks useful, then I add a second one, then a third, and somewhere around the fourth my agent starts behaving like a person who has read the entire menu out loud before ordering. It picks the wrong tool. It picks a tool that is almost right. It calls the filesystem server’s read_file when the built-in read_file was sitting right there. Nothing is broken, exactly. The thing just gets duller. The author explains that many coding agents include every MCP tool’s schema in every request, inflating prompts and causing the model to “pick from a bloated menu.” By inspecting Vercel’s fx (a Zig coding agent) and measuring real MCP-server payload sizes, they show fx avoids this by using a compile-time tool registry and a search/deferral mechanism: MCP tools are not actually included in the model-facing tool list at the start of a turn, but instead are discovered and inserted only after the model selects a specific “door” tool by name. This results in stable, bounded tool exposure (typically fifteen function schemas on a full turn, despite many installed MCP servers) with special handling for vision and provider-executed tools. The article also quantifies the cost (extra steps and a hard ceiling of five search results per capability query), compares the approach to OpenAI Codex’s runtime-based deferral and spec budgeting, and concludes that fx trades extra round trips and tighter discovery limits for a more reliable tool list size—avoiding the failure mode where the model has too many options and chooses wrong. Read the full blog for free on Medium. Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor. Published via Towards AI