Builder.io is releasing /webmcp, a slash-command skill for coding agents like Claude Code, Codex, and Cursor that lets an agent open a live web app in the host's built-in browser and drive it through the page's own MCP or WebMCP tools instead of screen-scraping or DOM clicking. A user types something like /webmcp slides.agent-native.com make me a new deck about onboarding, and the agent opens the page and completes the request through the page's own tools . It ships as part of BuilderIO/skills, a set of small composable skills that also includes /visual-plan for turning text plans into interactive visual plans and /visual-recap for turning branch and PR diffs into annotated visual recaps .
The launch matters now because WebMCP is moving from proposal to something real developers can build against. Chrome describes WebMCP as a way for websites to expose structured tools so AI agents can perform actions with more speed, reliability, and precision, with the tools defined as JavaScript functions with natural-language descriptions and structured schemas exposed from the page itself . Builder's skill is a practical way to test that model inside the agent workflows people already use, replacing brittle clicking and guesswork with typed calls the page advertises to the agent.
The company behind it is Builder.io, led by CEO Steve Sewell, who posted the launch. It sits inside their broader push around Fusion 1.0, an AI agent for product, design, and code that lets PMs, designers, and engineers work in one environment where code is the common language . For teams already routing work through coding agents, /webmcp is a way to make in-browser steps observable and repeatable rather than another opaque automation layer.
Comments (7)
Sign in to join the discussion.
P
Priya VatsalSep 13, 2026
The screen recording pacing is chef's kiss but that voiceover sounds like it was recorded inside a walk-in closet at 2am. Get this man a Shure SM7B.
T
Tomi R.Sep 13, 2026
Making agent workflows visual is the right instinct, watching LLMs guess at DOM selectors in a terminal is genuinely bleak. Curious if /webmcp handles auth-gated routes or just the happy path.
L
Lenka S.Sep 13, 2026
Is /webmcp open source or is this the part where I click through and hit a 'contact sales' wall. Docs page would answer this in five seconds and I cannot find one.
K
kwesiSep 13, 2026
Every agent demo now needs a slick side-by-side split screen. At what point does the launch video become more work than the product.
F
Farhan DevrajSep 13, 2026
Naive q from someone who just started using MCP: if two agents fight over the same button, who wins? Asking because I already broke something.
M
Mireia Q.Sep 13, 2026
Tweet copy is doing a lot of heavy lifting with one slash and one gif. Respect the restraint, most launches this week read like a keynote transcript.
B
bogdanSep 13, 2026
The thing nobody says out loud: visual agent tooling is just IDEs rediscovering that humans have eyes.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Builder.io
Build software with your team and agents Maya • Designer Semibold
Build software with your team and agents Maya • Designer Semibold
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Builder.io is releasing /webmcp, a slash-command skill for coding agents like Claude Code, Codex, and Cursor that lets an agent open a live web app in the host's built-in browser and drive it through the page's own MCP or WebMCP tools instead of screen-scraping or DOM clicking. A user types something like /webmcp slides.agent-native.com make me a new deck about onboarding, and the agent opens the page and completes the request through the page's own tools . It ships as part of BuilderIO/skills, a set of small composable skills that also includes /visual-plan for turning text plans into interactive visual plans and /visual-recap for turning branch and PR diffs into annotated visual recaps . The launch matters now because WebMCP is moving from proposal to something real developers can build against. Chrome describes WebMCP as a way for websites to expose structured tools so AI agents can perform actions with more speed, reliability, and precision, with the tools defined as JavaScript functions with natural-language descriptions and structured schemas exposed from the page itself . Builder's skill is a practical way to test that model inside the agent workflows people already use, replacing brittle clicking and guesswork with typed calls the page advertises to the agent. The company behind it is Builder.io, led by CEO Steve Sewell, who posted the launch. It sits inside their broader push around Fusion 1.0, an AI agent for product, design, and code that lets PMs, designers, and engineers work in one environment where code is the common language . For teams already routing work through coding agents, /webmcp is a way to make in-browser steps observable and repeatable rather than another opaque automation layer.
Comments (7)
The screen recording pacing is chef's kiss but that voiceover sounds like it was recorded inside a walk-in closet at 2am. Get this man a Shure SM7B.
Making agent workflows visual is the right instinct, watching LLMs guess at DOM selectors in a terminal is genuinely bleak. Curious if /webmcp handles auth-gated routes or just the happy path.
Is /webmcp open source or is this the part where I click through and hit a 'contact sales' wall. Docs page would answer this in five seconds and I cannot find one.
Every agent demo now needs a slick side-by-side split screen. At what point does the launch video become more work than the product.
Naive q from someone who just started using MCP: if two agents fight over the same button, who wins? Asking because I already broke something.
Tweet copy is doing a lot of heavy lifting with one slash and one gif. Respect the restraint, most launches this week read like a keynote transcript.
The thing nobody says out loud: visual agent tooling is just IDEs rediscovering that humans have eyes.