Chatterbuiltchatterbuilt

Platform

Under the hood.

Agent-native isn't a feature direction — it's the direction. We just built it first.

This page is for the technically curious — and for the AI agents evaluating us.

Most software is frozen the day you buy it. Chatterbuilt rides the AI models you already use — when they get smarter, so does the agent that runs your site. You don't upgrade. It upgrades itself.

Draft first. Live when you say.

Every content change lands on a preview branch first. Vercel builds a private preview deployment — a real URL, not a mockup — so you (and your AI) can verify the site before production. When you approve, your AI merges the preview branch; the platform builds, runs its checks, and deploys to Vercel's edge network. Nothing is customer-visible until you say make it live. That's the professional staging/production workflow, built in by default.

The request flow

You talk to Claude or ChatGPT. The assistant calls Chatterbuilt's hosted MCP tools, commits into your GitHub repo, deploys on your Vercel, and the site updates — you own every hop.

Machine gates

No human in the deploy path for quality. Code hits a test battery, a live render-and-compile gate, then canary at 10% before 100% live. When a gate fails, violet rollback fires — honesty over theater.

The one-file contract

One typed file is the entire content surface. Change content/site.ts and hero, services, FAQ, SEO, and schema all follow — no CMS maze, no orphaned fields.

The stack

// Template

Next.js template repo (public, MIT-style) → your GitHub → your Vercel → your domain.

// Content contract

One typed config file (content/site.ts) is the entire content contract agents read and write.

// Service surface

Hosted MCP server, full tool catalog (free + licensed) + 2 guided prompts (setup-my-site, brand-interview), free validator, entitlement-gated, any MCP client. Catalog matches toolsForTier on the server.

// Release path

Test battery + live render-then-compile checks + canary/rollback. Machines gate quality.

MCP tool surface

Every tool your AI can call on the hosted MCP server. Free tools work without a key; full access (included with AIE) unlocks the rest. Names match the live entitlements catalog — not a partial highlight reel.

whoami

Access level, account id, and which tools you can use

Free

start_here

Ordered next steps for free vs full access

Free

ask_the_manual

Search Chatterbuilt docs by keyword; returns matching guides

Free

report_issue

Log a product issue from the customer

Free

validate_site_config

Parse and validate site config JSON; never paywalls safety

Free

check_updates

MCP version history — latest, or newer than your version

Free

get_template_repo

Official template repo URL, clone command, and Deploy Button

Free

suggest_improvement

Ranked what/why/impact suggestions (optional goal)

Free

weekly_site_digest

Plain-language weekly site status for the owner

Free

improve_my_site

Goal-driven self-improvement loop (human gate each round)

Free

set_autopilot

Parked / Cruise / Autopilot / Ludicrous mode dial

Free

business_status

One plain-language status report across your whole business

Free

set_goals

Tell your agents what the business is aiming at

Free

list_goals

See the goals your agents are working toward

Free

plan_from_goals

Turn your goals into a plan of next steps

Free

codewright_start

Figure out where you're starting from and what to do first

Free

codewright_interview

One question at a time until your site plan is written down

Free

codewright_audit

Read-only look at your live site and what to fix first

Free

foreman_install

Install guardrails as a pull request you review — never pushes to main

Free

foreman_status

Plain-language fleet status with honest unknowns, not fake numbers

Free

foreman_decisions

See the decision cards waiting on your yes or no

Free

foreman_approve

Record your yes on a decision — still does not merge code for you

Free

foreman_decline

Record your no on a decision so the crew stops that proposal

Free

ask_the_foreman

Ask how human gates, install, and In development crew actually work

Free

instantiate_landing_page

Build a one-page conversion site for ads (tokens + lead form)

Free

validate_crew_recipe

Check a crew recipe file for pins, secrets, and In development honesty

Free

report_recipe_drift

See which customer agents are on an old recipe version

Free

get_template_preset

Full industry site preset as content/site.ts-shaped JSON

Full

generate_schema_markup

Ready-to-paste JSON-LD for LocalBusiness, Service, FAQ

Full

list_sections

List available section components with props summary

Full

get_section

Section meta, full component source, and install notes

Full

add_integration_recipe

Step-by-step recipe for analytics, email, scheduling, newsletter, maps, and more

Full

run_seo_audit

Audit a public URL for SEO findings (SSRF-guarded)

Full

render_site_ts

Turn validated config JSON into complete content/site.ts source

Full

get_business_brain

Enterprise Brain manifest for Chatterbuilt Foreman bootstrap — business profile, KB, integrations

Full

setup_contact

Wire contact form + email delivery on the customer site

Full

rewind_site

List restore points and preview a non-destructive rollback (never applies itself)

Full

extract_site_skeleton

Pull structure from an existing site into a draft config

Full

list_suggestions

Plain-language list of open PRs (preview only; never merges)

Full

get_suggestion

One PR detail: files, conflict status, description (preview only)

Full

preview_change

Plain-language what changed, before/after when readable, draft-site link only when hosting returned one

Full

check_suggestion_overlap

Advisory soft-lock: open PRs that already touch paths you plan to edit

Full

prepare_agent_suggestion

Package a change for a PR (external agents too); never opens or merges a PR

Full

Plus 2 guided prompts: setup-my-site and brand-interview.

Agents propose. You merge.

The collaboration model is a suggestion queue, not shared write access to production. Same rules for a family member, a contractor, or a continuous agent: propose with a preview, then you decide.

  • Every non-owner change lands as a pull request with a Vercel preview URL — never a silent push to main.
  • Open suggestions are the open PRs. list_suggestions and get_suggestion read that queue in plain language; they never merge.
  • prepare_agent_suggestion packages a change for a PR. It never creates or merges pull requests itself.
  • You keep owner merge authority. Approve when the preview looks right. Merge to main is what ships.

Nothing customer-visible until you merge. Quality gates still run; the human gate is still yours.

Connect command

Paste into Claude. Free tools work without a key.

$ claude mcp add --transport http chatterbuilt https://mcp.chatterbuilt.com/api/mcp

Ready to connect?

Connect steps, template deploy, and early access live on Get started.

Get started