--- title: Designing APIs for agents | daily.dev description: Webflow shares lessons from redesigning its APIs for AI agents using the Model Context Protocol (MCP). The core insight: developer APIs optimized for human... canonical: https://daily.dev/posts/designing-apis-for-agents-j6erh0xjr twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Designing APIs for agents | daily.dev og:description: Webflow shares lessons from redesigning its APIs for AI agents using the Model Context Protocol (MCP). The core insight: developer APIs optimized for human... og:url: https://daily.dev/posts/designing-apis-for-agents-j6erh0xjr og:image: https://api.daily.dev/og/posts/j6eRh0xJR.png og:image:alt: Designing APIs for agents og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Designing APIs for agents **[The New Stack](https://daily.dev/sources/newstack)** · 12 min read · 4 upvotes · 0 comments ## Summary Webflow shares lessons from redesigning its APIs for AI agents using the Model Context Protocol (MCP). The core insight: developer APIs optimized for human flexibility fail when exposed directly to agents, which need intent-driven, task-level interfaces with fewer dependent calls and clearer error guidance. Webflow evolved from wrapping existing endpoints to a layered tool architecture with domain-level groupings, workflow tools, and guide tools. They also explored filesystem-style abstractions where agents manipulate structured project representations as code. Infrastructure was built on Cloudflare Durable Objects for stateful sessions, with a WebSocket bridge for live Designer actions. Observability via MCPCat revealed unexpected usage patterns — canvas design accounted for 27.7% of sessions, a silent error affected 10%+ of sessions, and 58.7% of sessions clustered around just three workflows. A hosted MCP connector launched in early 2026 drove 6.7X session growth. The key takeaway: designing for agents means building full operating environments, not just exposing APIs. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why do agents struggle when developer REST APIs are directly wrapped as MCP tools? Developer APIs assume a human can read docs, manage state, and handle failures manually, but agents rely entirely on tool descriptions and schemas with no implicit context. Wrapping granular endpoints as MCP tools forces agents into many chained tool calls, increasing latency, token consumption, and failure probability, since the model must preserve intermediate identifiers and decide whether errors are retryable. _Teams building agent-facing tools track these MCP design patterns on daily.dev._ ### What is a layered tool architecture for scaling an MCP server as tool count grows? It organizes capabilities into domain tools that group broad product areas, actions that offer composable typed primitives within each domain, workflow tools that encapsulate orchestration-heavy tasks, and guide tools that supply operational instructions for safer execution. This structure gives agents a navigable map instead of exposing hundreds of standalone flat tools, reducing discovery and orchestration overhead as the product surface expands. _Engineers scaling MCP tool surfaces compare architecture patterns like this on daily.dev._ ### How much did session growth increase after launching a hosted MCP connector for Claude? Session growth increased 6.7x over three months after launching a hosted MCP connector starting with Claude in early 2026, which reduced setup friction by letting users connect through a managed integration instead of manually configuring MCP connections and authentication. Designing directly in the canvas through third-party agents also unexpectedly accounted for 27.7% of sessions. _Developers evaluating MCP connector adoption strategies follow real deployment data like this on daily.dev._ ## Similar posts on daily.dev - [How to Build AI Agents That Can Control Cloud Infrastructure](https://daily.dev/posts/how-to-build-ai-agents-that-can-control-cloud-infrastructure-yyzbrnqg9) · freeCodeCamp · 2 upvotes · 0 comments - [Bridging the Gap - How We Made AI Agents 10x Developers in Our Organization](https://daily.dev/posts/bridging-the-gap---how-we-made-ai-agents-10x-developers-in-our-organization-9xldqiaaq) · DEV · 4 upvotes · 0 comments - [Why MCP Shouldn’t Wrap an API One-to-One](https://daily.dev/posts/why-mcp-shouldn-t-wrap-an-api-one-to-one-m1d3gzxtr) · Nordic APIs · 2 upvotes · 0 comments - [AI Agents Are Growing an Infrastructure Layer](https://daily.dev/posts/ai-agents-are-growing-an-infrastructure-layer-jk8bjcch7) · Medium · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#architecture](https://daily.dev/tags/architecture), [#observability](https://daily.dev/tags/observability), [#mcp](https://daily.dev/tags/mcp), [#nocode](https://daily.dev/tags/nocode) [View this post on daily.dev](https://daily.dev/posts/designing-apis-for-agents-j6erh0xjr) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Designing APIs for agents","url":"https://daily.dev/posts/designing-apis-for-agents-j6erh0xjr","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/designing-apis-for-agents-j6erh0xjr"},"datePublished":"2026-08-01T14:06:41.547Z","dateModified":"2026-09-14T07:15:51.078Z","description":"Webflow shares lessons from redesigning its APIs for AI agents using the Model Context Protocol (MCP). The core insight: developer APIs optimized for human...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/16a0ddaea5d238b55c3acd9a01152a0d?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/16a0ddaea5d238b55c3acd9a01152a0d?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The New Stack","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The New Stack","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/newstack","url":"https://daily.dev/sources/newstack"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/designing-apis-for-agents-j6erh0xjr","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":4},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,architecture,observability,mcp,nocode","timeRequired":"PT12M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The New Stack","item":"https://daily.dev/sources/newstack"},{"@type":"ListItem","position":3,"name":"Designing APIs for agents"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/designing-apis-for-agents-j6erh0xjr#faq","mainEntity":[{"@type":"Question","name":"Why do agents struggle when developer REST APIs are directly wrapped as MCP tools?","acceptedAnswer":{"@type":"Answer","text":"Developer APIs assume a human can read docs, manage state, and handle failures manually, but agents rely entirely on tool descriptions and schemas with no implicit context. Wrapping granular endpoints as MCP tools forces agents into many chained tool calls, increasing latency, token consumption, and failure probability, since the model must preserve intermediate identifiers and decide whether errors are retryable. Teams building agent-facing tools track these MCP design patterns on daily.dev."}},{"@type":"Question","name":"What is a layered tool architecture for scaling an MCP server as tool count grows?","acceptedAnswer":{"@type":"Answer","text":"It organizes capabilities into domain tools that group broad product areas, actions that offer composable typed primitives within each domain, workflow tools that encapsulate orchestration-heavy tasks, and guide tools that supply operational instructions for safer execution. This structure gives agents a navigable map instead of exposing hundreds of standalone flat tools, reducing discovery and orchestration overhead as the product surface expands. Engineers scaling MCP tool surfaces compare architecture patterns like this on daily.dev."}},{"@type":"Question","name":"How much did session growth increase after launching a hosted MCP connector for Claude?","acceptedAnswer":{"@type":"Answer","text":"Session growth increased 6.7x over three months after launching a hosted MCP connector starting with Claude in early 2026, which reduced setup friction by letting users connect through a managed integration instead of manually configuring MCP connections and authentication. Designing directly in the canvas through third-party agents also unexpectedly accounted for 27.7% of sessions. Developers evaluating MCP connector adoption strategies follow real deployment data like this on daily.dev."}}]} ```