NewToolJet MCP · OPEN SOURCE · MIT LICENSED

Your coding agent can now
build inside ToolJet.

Install the ToolJet plugin in Claude Code, Codex, or GitHub Copilot, and it builds real ToolJet apps - pages, queries, components, and permissions. The result is native, visually editable, and governed from the first prompt.

Get the plugin See how it works

Works with any standards-compatible MCP client

ToolJet MCP session building an incident dashboard app

Bring the coding agent your team already uses

Codex

Claude Code

Cursor

GitHub CopilotAny MCP client

One protocol. The full app model.

More than code generation

ToolJet MCP gives coding agents structured access to what makes an internal app work - interface, data, logic, permissions, and release state. Each is backed by ToolJet's own governed contracts, not a guessed config key. Every change is dry-run, applied, then checked against what actually persisted.

get_datasource_query_schema

Reads before it writes

Pulls real datasource and query contracts before writing a single query - no guessed configuration keys.

create_app

Builds native ToolJet apps

Real pages and components, wired together - not detached markup your team has to translate by hand.

create_tables

Models ToolJet DB tables

Creates and evolves tables, seeds rows against the real sequence, and generates a matching Form from the schema.

add_events

Wires real behavior

Click handlers, success/failure flows, and query bindings - attached the same way the visual builder attaches them.

lint_app_spec

Plans, applies, then checks

Dry-runs a phase, applies it with a one-time plan token, then statically validates what actually got persisted.

manage_app_permissions

Administers, not just builds

Invites users and restricts page, query, or component access - every mutation requires explicit confirmation.

From prompt to production

Stay in your agent.
Keep ToolJet in control.

Connect

Generate a personal access token in ToolJet, then add the MCP server to your coding agent.

Prompt

Describe the app in plain language - the agent finds your data sources, then builds and wires it in one pass.

Review

It hands back an editor and a viewer link. Every component, query, and event stays visually editable.

Verify + ship

Every change is dry-run, applied, then checked - before anything leaves the development environment.

Native by default

The agent builds what your team can maintain.

MCP writes to ToolJet’s application model. When the agent is done, your team can select the same table, inspect the same query, and change the same event from the visual editor.

Real ToolJet components - not screenshots or generated markup
Managed queries and datasource bindings, never hardcoded credentials
Visual edits and agent edits share one version history
Custom code only when the app actually needs it
Runs on your AI client's own subscription - doesn't draw down ToolJet AI credits

support-operationsDraft v4

SupportIncident overview

Open147−12% today

Response time18mWithin target

Edited from Codex via MCP Open in visual editor

Governance is part of the protocolExplore enterprise security

Give agents useful access. Keep the boundaries.

The agent gets the tools it needs, without becoming a back door around your policies. The MCP server talks only to your ToolJet instance - but your AI client's model provider still sees the prompts and data it reads.

SOC 2ISO 27001GDPRSelf-host

User-scoped access

The MCP session inherits every permission the connecting developer has across the workspace.

A complete audit trail

Every change is attributed to the agent session and the person who initiated it.

Credentials stay in ToolJet

Agents use governed resources without ever receiving database or API credentials.

Human-controlled release

Agent work stays in the development environment until it passes your normal promotion process.

Explore enterprise security