Tilda editor MCP server
Tilda MCP
A cautious MCP server for the Tilda editor that plans changes, rereads after writing and can roll back, instead of clicking the interface directly
Medium risk
We rate an entry medium when the tool runs code, makes network calls or reads project files. Check what exactly it does before installing.
Why this level
- Writes changes to real site pages through an authorized editor session
- Publishing and unpublishing are separated into their own confirmable actions
Install
Manual install
pnpm install --frozen-lockfile && pnpm buildInstalls dependencies and builds the project from source.
This is third-party code. Review the repository files before installing.
What it does
Tilda has no official API for editing page content, so the server works through an authorized editor session in a browser with remote Chrome debugging. Instead of arbitrary clicks and JavaScript, it exposes fourteen tools: status, a capability list, one-time task authorization scoped to specific targets, audit, a change plan, apply, result verification and rollback, separate publish and unpublish actions, and a fixed recipe for common page edits. Every change goes through a dry-run plan, a real write only on an explicit request, a reread to compare against the expected outcome, and restoration if the outcome is not confirmed.
Who it is for. For people running Tilda sites who want an agent to make targeted page edits with verification and rollback.
Good fit when
- You need targeted, verifiable edits to text, SEO fields or a page's HEAD code on Tilda
- You need every change to be reversible if the outcome does not match expectations
- You need publishing and editing to be controlled separately
Not a fit when
- You need image uploads, bulk catalog edits, form handling or arbitrary nested HTML: explicitly out of scope for v1
- You want a stable finished product: this is a 1.0.0 pre-alpha release with public code and no real credentials included
- You do not have your own dedicated Chrome profile with remote debugging for an authorized session
Example request
Check the current heading on my Tilda landing page, propose a change plan and apply it only after I confirmLimitations
The project has a single author and is in public pre-release status, with no ready list of real projects and pages: it must be built privately by the user. It needs Node.js 20+, pnpm, and a dedicated Chrome profile with remote debugging on a loopback address. It does not claim to cover every Tilda editor feature and is not an official Tilda product.
How to disable. Stop the MCP server process, remove it from your client config, and close the dedicated Chrome remote-debugging profile.
MCP
- Transport
- stdio
- Authentication
- not required
Security check
- Writes changes to real site pages through an authorized editor session
- Publishing and unpublishing are separated into their own confirmable actions
README in short
The README describes the safety architecture in detail: single-task authorization scoped to exact targets, browser-session binding, state snapshots, comparison hashes, post-write rereads, rollback and separately gated publication. It lists the version 1.0 boundaries explicitly: what is unsupported, such as file uploads and bulk catalog operations. Quick-start commands and environment requirements are included. Apache-2.0 license; the project is not affiliated with Tilda.
FAQ
Does the public repository include real account data?
No, the public checkout contains only generic code, synthetic tests and safety documentation, with no project IDs, page IDs or credentials.
What happens if a change is not confirmed after writing?
The server rereads the target, compares it to the expected outcome, and restores the prior state or rolls back if there is no proof the requested result was achieved.
Related
Official Claude Code plugin from the Next.js repo: adopting Cache Components and Partial Prefetching, and verifying changes on a running dev server
Official shadcn/ui skill and MCP server: the agent searches registries, adds components via the CLI and follows the library's rules
A frontend design skill with 23 commands, live browser iteration and 61 detector rules for common AI-generated UI tells
Vercel Agent Skills
Agent Skills
Vercel's skills: React and Next.js performance, UI guideline audits, React Native, component composition and deploys