Wiring up the roof extraction AI workflow inside a Claude Plugin. Slash Commands and Skills for simple measurements or special modes like roofing
MCP App drops UI right inside the chat window, user types /roof 123 Main st and is presented with the UI for immediate measuring and editing.
On completion, the agent picks up the response and pairs with a user customizable Skill file (for pricing, materials, report generation, whatever post estimate workflow they want - simply customized with the Skill text file)
Could pair well with Claude Small Business plugin for roofers, fence installers, landscapers, power washers etc
User AI agents (like Claude Cowork) outside of SaaS walls are going to want access to all workflows, existing APIs will help but when it comes to user collaboration with the agent, existing UX might need to change.
@claudeai@ClaudeDevs@grok will MCP Apps be added to Grok Build?
I agree, a fast web UI for humans is still needed, and then MCP/API+Skills for agents. A third intermediate option is an MCP App for your interactive map UI directly in the agent chat, allows the user to make focused decisions and then hand off curated JSON data to the agent to act on via a skill. What I like about this workflow is giving the local agent (claude, codex, grok) deterministic information that it can then act on with code, reason and other connectors. A Skill with API access for the agent is similar but the MCP App gives the human some control in the flow
Here's a roofing estimate example where the user can check the measurements before the agent takes over and creates the customer and sends an estimate
https://t.co/H5vgIv7xzk
This is the first end-to-end vertical workflow I've seen someone demonstrate on CFSB. The measurement-to-estimate-to-QBO chain is exactly the kind of thing that makes the platform real for a specific industry. Where did you find the show-measurement MCP app — is that in the Cowork directory or did you build it?
CFSB integration - It's just a demo right now but it does work pretty well, more importantly I think it shows where things could be heading: AI embedded in a monolithic SaaS or "bring your own" local AI company brain connecting to smaller services. I see pros and cons for both tbh.
show-measurements is a custom tool I built into a larger project but after talking to a few users, I'm going to break it out into something more focused for trade estimations. Waitlist here if you wanted to eventually try it out, appreciate all feedback: https://t.co/DhzAhGHnVR
Claude for Small Business plugin integration -
This process could use some polish but you can definitely go from site address measurements/roof analysis to a new Customer and Estimate in Quickbooks, all right from the Claude Cowork (about 3mins end to end)
Workflow from Claude:
Measure — show-measurement (MCP App) with mode: "roofing", trace the roof, click Done → roofing envelope
Price — roofing-estimating Skill turns it into a line-item estimate using your saved rates and custom configurations
SBS connectors: QuickBooks → create the customer + an Estimate object with the line items (squares, underlayment, tear-off, labor, etc.).
@ClaudeDevs@QuickBooks
@claudeai great timing! working with some users to automate Cowork Small Business plugin with roofing and fence installation estimates
https://t.co/uLE3jBj78O
Claude for Small Business plugin integration -
This process could use some polish but you can definitely go from site address measurements/roof analysis to a new Customer and Estimate in Quickbooks, all right from the Claude Cowork (about 3mins end to end)
Workflow from Claude:
Measure — show-measurement (MCP App) with mode: "roofing", trace the roof, click Done → roofing envelope
Price — roofing-estimating Skill turns it into a line-item estimate using your saved rates and custom configurations
SBS connectors: QuickBooks → create the customer + an Estimate object with the line items (squares, underlayment, tear-off, labor, etc.).
@ClaudeDevs@QuickBooks
I agree and I think we'll see vertical SaaS needing to allow direct agent access to the tools in their stack instead of (or in addition to) having their own chat agents they run for you. There is just too much domain knowledge in the trades and you might be better off building from your own experience with something like Claude if you want AI estimates. Skill files can help a lot with encoding that knowledge, Claude will even help you write them
Wiring up the roof extraction AI workflow inside a Claude Plugin. Slash Commands and Skills for simple measurements or special modes like roofing
MCP App drops UI right inside the chat window, user types /roof 123 Main st and is presented with the UI for immediate measuring and editing.
On completion, the agent picks up the response and pairs with a user customizable Skill file (for pricing, materials, report generation, whatever post estimate workflow they want - simply customized with the Skill text file)
Could pair well with Claude Small Business plugin for roofers, fence installers, landscapers, power washers etc
User AI agents (like Claude Cowork) outside of SaaS walls are going to want access to all workflows, existing APIs will help but when it comes to user collaboration with the agent, existing UX might need to change.
@claudeai@ClaudeDevs@grok will MCP Apps be added to Grok Build?
These are great, definitely can eat tokens but well worth it IMO. Works great from VSCode+ClaudeCode too
"Workflow launched in background. Task ID: wq7f4wtu8
Summary: Adversarial review of the waitlist-approval magic-link onboarding diff: correctness, security/abuse, React edge cases — each finding independently verified"
"The review confirmed 9 findings — the three HIGH ones are the same real bug caught independently by 3 reviewers"
I love conversations like this.
A goat farmer is building his own software to manage his land clearing and goat grazing business.
The future of vertical saas belongs to the builders!
agents orchestrating agents is all over my feed today. i'm a master electrician, not a coder. but i already run AI agents that read my notes before they act and put my electrical knowledge in a homeowner's hands. the trades are going to feel this harder than anyone expects.
Part of what makes this work is by using AI to selectively remove the tree canopy when necessary, giving the rest of the pipeline access to additional data for cross checks. In Claude, the user can also see both images and make any adjustments to the roof edges.
Working on an agentic workflow for roof edge/facet extraction from a single satellite image from the Google Solar API. It involves multiple AI calls and deterministic code and can produce 95% accurate edge/facets results for any US address (even when foliage covers a portion of the roof). Was stuck on some of it and decided to use Grok Build to iterate on a plan, then used Claude Opus 4.8 to complete the deterministic code and pipeline.
Still a work in progress but good to remember that different agents bring different answers, a lot like how humans work in groups.
@toddsaunders I had some good luck with Opus 4.8 today (actually following a plan that Grok Build created), I think its more thoughtful then 4.7
https://t.co/H5LuIhbx6F
Working on an agentic workflow for roof edge/facet extraction from a single satellite image from the Google Solar API. It involves multiple AI calls and deterministic code and can produce 95% accurate edge/facets results for any US address (even when foliage covers a portion of the roof). Was stuck on some of it and decided to use Grok Build to iterate on a plan, then used Claude Opus 4.8 to complete the deterministic code and pipeline.
Still a work in progress but good to remember that different agents bring different answers, a lot like how humans work in groups.
Working on an agentic workflow for roof edge/facet extraction from a single satellite image from the Google Solar API. It involves multiple AI calls and deterministic code and can produce 95% accurate edge/facets results for any US address (even when foliage covers a portion of the roof). Was stuck on some of it and decided to use Grok Build to iterate on a plan, then used Claude Opus 4.8 to complete the deterministic code and pipeline.
Still a work in progress but good to remember that different agents bring different answers, a lot like how humans work in groups.
Welcome to Blue Collar Builders!
Cory LaChance inspired me to start a series spotlighting folks in the trade who are building software using AI.
Cory normally works with chemical plants and refineries, but now he's building AI software for his company.... with no pervious experience writing code.
He built a full agentic application that industrial contractors are using every day.
It reads isometric drawings and automatically extracts every weld count, every material spec, every commodity code.
My favorite thing he said was, "I did this with zero outside help other than the AI. My favorite tools are screenshots, step by step instructions, and asking Claude to explain things like I'm five."
I hope you enjoy this episode as much as I did.
And I can't wait to meet more Blue Collar Builders.