MCP App Store
by KeenEthics
Apper - Website & App Builder icon

Apper - Website & App Builder MCP Server for Claude and ChatGPT

By CompanyHub IT Solutions Pvt Ltd

Overview

Apper lets you build real, working apps and websites using plain language. Describe what you want, and Apper creates it with a real database, secure server-side logic, and a live preview you can refine right from your conversation. Security is built in. Your API keys never pass through the chat, and access rules and permissions protect your data. Perfect for founders validating ideas, teams building internal tools, or anyone with a project and no dev team.

How to connect

  1. Claude: Open the Apper - Website & App Builder listing in the Claude connectors directory and choose Connect. Claude listing Apper - Website & App Builder exposes 32 tools here.
  2. ChatGPT: Open the Apper - Website & App Builder listing in the ChatGPT app directory and choose Connect. ChatGPT listing Apper - Website & App Builder exposes 32 tools here.

Tools

apply_patch

ChatGPT
Edit existing files with unified diffs (git-diff format). Pass each file's sha from read_files as baseVersion so a stale read is caught as a conflict. A hunk whose context no longer matches is skipped, not fatal; rebuild it from failedHunks' contextBefore/contextAfter or finalContent — never from memory. Not for new files.

check_mcp_version

ChatGPT
Checks whether this connector is up to date. Call once near the start of a session with the connector's own version as clientVersion. If upToDate is false, relay the returned message to the user verbatim and ask them to reconnect the Apper MCP server.

connect_database

ChatGPT
Switch the project from mock data to a live Apper database, provisioning the schema (derived from the mock JSONs), RLS, profiles, and permissions. One-time and irreversible — it deletes the mock layer. The schema, RLS, and permissions themselves stay editable via update_database. Requires the user's explicit approval of a plain summary of tables, profiles, and access first, and never runs in the same turn as create_app. Read get_rls_policy_instructions before building the rls payload. The platform injects the env keys — never ask the user for them.

create_app

ChatGPT
Create a new Apper project and return its projectId. Read get_create_app_instructions and get_design_directives first. The seeded src/pages/Home.jsx is a placeholder with no route export — / is a 404 until you overwrite it.

create_database_components

ChatGPT
Add Table | Field | RLS | Profile | Permission components to the connected database, batched into one call per approved change, then update the frontend to match. Never changes or removes anything that exists. A new table carries name, label, plural, with its fields/RLS/permissions nested in children[]; a field, RLS policy or permission on an existing table carries that table's numeric tableId (from get_backend_schema) in properties. Shapes the backend rejects otherwise: every component type is exactly Table|Field|RLS|Profile|Permission with the field's data type in properties.type; Lookups name their target table; access tokens are Full|Read|Create|Update|Delete|Inaccessible (comma combos allowed — 'Inaccessible' not 'None', 'Create' not 'Insert'); re-creating an existing profile fails, and a public profile is never the default. Read get_rls_policy_instructions before any RLS component.

create_edge_function

ChatGPT
Create and deploy a server-side function (secret-key API calls, email, payments, importing or validating data via sdk — not plain page CRUD). The only way to create one; writing the file yourself produces a dead file. Deploys the endpoint, syncs the repo copy, and writes the VITE_<NAME> env var the frontend calls it by. Source shape and call site are in the platform contract.

create_secrets

ChatGPT
Register secret names and get the dashboard URL where the user enters the values. You never see values. Confirm the user has entered a secret before relying on a function that reads it. If a user pastes a live key into chat, tell them to rotate it.

delete_database_components

ChatGPT
Remove Table | Field | RLS | Profile | Permission components from the connected database, batched into one call per approved change, then update the frontend to match. Each component carries its real id from get_backend_schema and empty properties, except a Field delete, which also carries its tableId. Deleting a table removes its data.

delete_edge_function

ChatGPT
Delete an edge function and its repo copy. Remove call sites and the dead VITE_<NAME> reference in the same change.

delete_files

ChatGPT
Delete files in one commit. Never a platform-owned path.

delete_secret

ChatGPT
Delete a secret by its numeric id from list_secrets.

generate_message_id

ChatGPT
Call this for every new user message, before any other tool EXCEPT the one that establishes projectId: for a brand-new app, call create_app first (no projectId exists yet); for an existing app whose projectId isn't already known in context, call search_apps first. Once you have a projectId, call this tool with it to record the user's message and get back a fresh messageId (an integer) that represents that one user message — every other tool call made while handling this message must pass this same messageId. When the user sends their NEXT message, call this again for a new messageId; never reuse a messageId across two different user messages. Exactly one call per user message: never call this a second time while still handling the same message, no matter how much work it involves or how many tool calls it takes. A multi-part request, a mid-task re-orientation, and a recovery from any failure — a failed edit, a failed build, an auth error — are all still the same user message, so they keep the same messageId.

get_backend_schema

ChatGPT
The source of truth for mock vs connected: a schema means connected, empty/error means mock. Also the only source of the numeric ids update_database needs.

get_backend_schema_filtered

ChatGPT
Same as get_backend_schema, scoped to the tables you name (with or without _c) plus their RLS policies. Profiles always return in full. Use it once you know which table you're changing.

get_build_status

ChatGPT
Check the build after a shouldBuild:true commit. Blocks server-side until a terminal status or ~75s — call it once, not in a loop. COMPLETED → preview_app. *-FAILED → read the error, fix the cause, commit again. Any other status → call once more.

get_create_app_instructions

ChatGPT
The platform contract and build guidance for creating a new Apper app. Read once, before create_app.

get_design_directives

ChatGPT
How to turn a spec into a complete, distinctly designed product: the product brief to write first, the design decisions to derive from it, and the quality and completeness floors every screen clears — plus a design draw picked by the server for this app (shell, type, palette, signature) so it does not look like the last one. Call it once per new app, before writing files, and before adding any new page or surface to an existing one.

get_edge_function

ChatGPT
Read one edge function's source by id. The repo copy lags — this is the source of truth.

get_edit_app_instructions

ChatGPT
The platform contract and edit guidance for changing an existing Apper app. Read once per session, before the first edit.

get_env_keys

ChatGPT
List env variable key names (values are never returned). Secrets live in create_secrets.

get_project_tree

ChatGPT
The project's full file tree. Read it once per session before editing; a path missing from it does not exist. Not needed right after create_app.

get_rls_policy_instructions

ChatGPT
How to translate .policy.js files into the JSON RLS policies connect_database and create_database_components / update_database_components / delete_database_components accept. Not part of the create/edit instructions — read it once, right before the first call that carries an rls payload.

list_edge_functions

ChatGPT
List the project's edge functions: id, name, executionUrl, isDeployed.

list_secrets

ChatGPT
List secret names (never values). Check before creating one.

patch_files

ChatGPT
Edit existing files with {oldStr, newStr} replacements. oldStr is copied verbatim from read_files output and must occur exactly once (line endings normalized). A non-matching edit is skipped, not fatal — everything else still commits; failedEdits lists the skips and finalContent gives each patched file's current content for a retry. Not for new files.

preview_app

ChatGPT
Return the app's preview URL. Only after get_build_status returns COMPLETED. Returns previewUrl, mock (true while the app runs on mock data) and, on mock, demoAccounts read from src/services/mockData/users.json (profile, email, password per user). Your reply ALWAYS contains the full previewUrl as plain text and, on mock, every demoAccounts entry one line each plus the "Connect to database" closing line — every time this tool runs, on create, edit and every phase. A warning or usersWarning in the result means the app is not finished: fix it, build, preview again.

read_files

ChatGPT
Read one or more files in a single call. Always read a file before patching or rewriting it — write_files replaces files whole, so writing from memory drops whatever you forgot. Returns each file's content and sha.

search_apps

ChatGPT
Find the user's Apper projects by projectId, name, subdomain, label, or preview URL. Omit search to list all. Every other tool needs a projectId from here or from create_app.

set_env_key

ChatGPT
Set a public VITE_* env variable. Rejects secret-looking keys. Edge-function variables are written by create_edge_function, not here.

update_database_components

ChatGPT
Change existing Table | Field | RLS | Profile | Permission components in the connected database, batched into one call per approved change, then update the frontend to match. Each component carries its real id from get_backend_schema and ONLY the properties that change; a Field or RLS update also carries its tableId. A Permission is identified by profileName (bare, e.g. 'Member') plus tableName, with id 0. Access tokens are Full|Read|Create|Update|Delete|Inaccessible. Read get_rls_policy_instructions before any RLS component. Nothing is created here — use create_database_components.

update_edge_function

ChatGPT
Update an edge function by id; send only changed fields. A rename writes the new env var — update every call site.

write_files

ChatGPT
Commit complete file contents. shouldBuild:true = commit, build, deploy, and a new version in the user's history — exactly one true per user message, on the commit that completes it; every other commit in the turn is false (commit only). commitMessage names the files and the change, and the true commit's message covers the whole turn. The only way to create a new file. Send formatted source — one statement or JSX element per line; a file with no line structure cannot be patched reliably and comes back as a warning. Everything commits except a mock-data JSON whose structure is wrong: it is left out and returned under errors with what is wrong and the expected shape — resend only that file, corrected. The result also reports autoFixes (repairs already made — nothing to do), warnings (will break the build, a screen, or a later edit — fix in the next commit), notes (quality — no tool call now; apply when you next touch the file), skippedFiles (frozen platform files, never committed) and a notice when the app as a whole has a problem (dead links, entities with no create/edit/delete, tables with no policy file, a placeholder Home).

apply_patch

Claude

check_mcp_version

Claude

connect_database

Claude

create_app

Claude

create_database_components

Claude

create_edge_function

Claude

create_secrets

Claude

delete_database_components

Claude

delete_edge_function

Claude

delete_files

Claude

delete_secret

Claude

generate_message_id

Claude

get_backend_schema

Claude

get_backend_schema_filtered

Claude

get_build_status

Claude

get_create_app_instructions

Claude

get_design_directives

Claude

get_edge_function

Claude

get_edit_app_instructions

Claude

get_env_keys

Claude

get_project_tree

Claude

get_rls_policy_instructions

Claude

list_edge_functions

Claude

list_secrets

Claude

patch_files

Claude

preview_app

Claude

read_files

Claude

search_apps

Claude

set_env_key

Claude

update_database_components

Claude

update_edge_function

Claude

write_files

Claude

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/apper-app-website-builder"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/apper-app-website-builder)