Appearance
Tips and guidelines
Tool calls run against your real project, so a few habits go a long way. Here is what we recommend when you build with an external agent over MCP.
Name your project first
The agent calls getMe to discover your workspaces and projects, then targets the project you named on each call. There is no persistent "active project" to switch, so check it picked the right one before letting it make changes.
Set up backend work before interface work
When a request spans both, have the agent create tables and backend workflows before building the interface that depends on them. That way the pages it builds can be bound to data that already exists.
Review as you go
Because tools run against your live project, keep an eye on changes and use the WeWeb editor and Preview to verify results.
Use a model that follows the tooling
For the best results today, we recommend Claude Opus as the underlying model. In our testing, it follows WeWeb tooling more reliably and produces higher-quality output than other models.
Let the agent read the docs
The agent can call getWeWebDocumentation and getIntegrationSkill to look up how WeWeb works before it builds. If a result is off, asking it to check the documentation for that feature often fixes the next attempt.
Send us feedback
Your feedback shapes what comes next. Ask the agent to use wewebFeedback to send a note about a tool or a piece of documentation directly to the WeWeb team.
CONTINUE LEARNING
Want to see MCP applied to a full build? Follow along as an AI agent builds a CRM from start to finish.