Raul Delapena Setiawan
In the rapidly expanding ecosystem of AI-driven development, the sheer volume of "coding agents" has reached a state of chaotic saturation. Every week brings a new terminal-based assistant promising to revolutionize software engineering. However, Vercel Labs has recently introduced a new entrant—fx—that manages to cut through the noise, not necessarily by promising superior code generation, but by setting a new, radical standard for technical transparency.
The Arrival of fx: A New Native Powerhouse
At its core, fx is a compact, high-performance native binary. Weighing in at approximately 6MB and written in the memory-safe Zig language, the project is currently in its early stages (v0.0.13) and remains Apache-2.0 licensed. It is explicitly experimental, with a bold disclaimer warning users that its architecture is subject to frequent, breaking changes.
Despite its diminutive size, fx is packed with modern features that signal the current trajectory of autonomous coding tools. It is entirely model-agnostic, allowing developers to plug in their preferred intelligence providers—whether that be a ChatGPT or Grok subscription, an endpoint routed through Vercel’s AI Gateway, or a locally hosted model via Ollama. By utilizing AGENTS.md files for directory-specific instructions and SKILL.md folders for modular agent skills, fx mirrors the industry’s move toward highly contextual, persistent coding environments.
The Transparency Paradigm: A Manifesto for MCP
What truly distinguishes fx from the dozens of other agents currently fighting for terminal space is its documentation. Specifically, the project publishes a comprehensive, granular breakdown of its implementation of the Model Context Protocol (MCP).
In an industry where technical documentation is often treated as an afterthought—or worse, a marketing brochure—fx takes the unusual step of providing a definitive "Support Matrix." The documentation explicitly lists which MCP features are operational (such as stdio, Streamable HTTP, resource templates, and JSON Schema 2020-12 support) and, more importantly, which features are missing.
The "Negative" Documentation
The project clearly states that it does not currently implement MCP Tasks, MCP Apps, Skills over MCP, or Enterprise-Managed Authorization. It does not attempt to act as an MCP server, nor does it install servers from a central registry.
For the average user, a list of what a piece of software cannot do might look like a weakness. For the professional ecosystem, it is a revelation. By defining the "edges" of its capability, fx allows API providers and developers to immediately understand what will function and what will be ignored. This honesty prevents the "black box" frustration common in AI tooling, where developers waste hours troubleshooting features that were never actually supported in the first place.
Chronology: The Evolution of Agent Interoperability
The need for this level of transparency is not merely a preference; it is a reaction to the last eighteen months of fragmented AI development.
- Early 2024: The explosion of LLM-based coding assistants leads to a fragmented landscape where every agent acts as a siloed island.
- Mid-2024: The Model Context Protocol (MCP) begins to gain traction as a way to standardize how AI models interact with data sources and development tools.
- Late 2024: Developers begin to realize that "MCP support" is being used as a catch-all marketing term. Tooling developers begin to implement inconsistent slices of the protocol, leading to widespread interoperability issues.
- October 2026: Measurements of OpenAPI 3.2 tooling indicate that the majority of modern AI-adjacent software lacks clear documentation regarding specification coverage.
- Present Day: Vercel Labs releases
fx, becoming the first major agent to codify its MCP coverage in a public, readable format, effectively challenging other vendors to do the same.
Supporting Data: The Cost of Ambiguity
The frustration surrounding the current state of AI tooling is best illustrated by the difficulty of measuring protocol support. In recent audits of the OpenAPI 3.2 landscape, the primary obstacle was not the complexity of the specification itself, but the lack of vendor transparency. Most tools fail to state their support levels, forcing users to reverse-engineer capabilities through GitHub issue threads and obscure changelogs.
The MCP landscape is currently trending toward this same abyss. Because the specification is moving rapidly, different clients are implementing different subsets of the protocol. Without a "coverage table"—a simple document detailing what is supported—interoperability remains a theoretical dream rather than a practical reality. fx serves as a proof-of-concept that this documentation is the cheapest and most effective piece of interoperability work a team can undertake.
Official Responses and Strategic Design
While Vercel has not released a formal "manifesto" beyond their documentation, the design choices within fx speak volumes about their development philosophy. Two features, in particular, highlight a focus on safety and integration:
1. The Hierarchical Permission Model
Security is the Achilles’ heel of autonomous agents. fx addresses this with a tiered approach to execution. Routine tasks—such as file reading or standard project builds—are permitted by default. However, any interaction that involves an MCP tool call is subjected to a "review model." This creates a human-in-the-loop or agent-in-the-loop verification process, ensuring that external API calls are vetted before execution. This makes the tool suitable for professional environments where "agent-runaway" is a significant concern.
2. Libfx: The Portable Runtime
Beyond the CLI, Vercel has exposed the core logic of the agent as libfx, a JavaScript library. This allows the runtime to function natively in Node.js or within the browser via WebAssembly. By making the agent runtime a library, Vercel is encouraging developers to build their own tools on top of the fx engine, rather than forcing them to work within the confines of a rigid, proprietary platform.
Implications for the Industry
The introduction of fx creates a new benchmark for the "Developer Experience" of AI agents. If the industry adopts this transparency-first approach, we can expect several immediate shifts:
- Standardized Interoperability: If every MCP client provides a coverage matrix, API providers can build servers with the confidence that their features will be utilized correctly by the client.
- Vendor Accountability: The "fx-style" transparency forces vendors to justify their gaps. If an agent claims to be "enterprise-ready" but fails to implement critical security protocols, the public documentation will make that failure transparent.
- Better Procurement Decisions: For engineering managers, evaluating agents will move from "which marketing site looks better" to "which tool supports the specific protocols our stack requires."
The Path Forward: A Call for Documentation
The list of "what we do not support" is not a gap to be hidden—it is a finding to be shared. By placing fx in the tools.apievangelist.com directory under the role of an active, credential-using agent, the community has acknowledged that the threshold for entry into the "professional" category has been raised.
The challenge now lies with the rest of the developer tooling ecosystem. If a project is built on open standards like MCP, there is no longer a valid excuse for obscuring the level of implementation. We need more projects to treat their protocol coverage as a primary piece of documentation.
If you are currently building an MCP client, the fx protocol page should be your template. If you are an engineering lead evaluating coding agents, your first question to the vendor should not be "does it work with MCP?" but rather, "can you provide the paragraph that says exactly what you do not support?"
Vercel has shown that transparency is not just an ethical choice—it is a competitive advantage that fosters trust in a crowded and skeptical market. The bar has been set; it is now up to the rest of the industry to clear it.
Ammar Sabilarrohman
Layla Zulfa
