Connect Webase to AI Tools

Webase exposes three MCP endpoints for different agent environments. Pick the one that matches your tool, configure it once, and your AI host can list projects, build apps, define data models, evaluate, and deploy — all from natural-language prompts. New to Webase MCP? Read the Overview first.

Claude Code (CLI)

Claude Code is a desktop coding agent — use the external endpoint so it gets upload_build and won't accidentally write source files via MCP.

claude mcp add webase \
  --transport http \
  https://www.webase.com/mcp/external/messages

Claude Code will open the OAuth flow in your browser the first time you call a tool.

Claude Desktop / Claude.ai

Claude Desktop / Claude.ai is a chat agent — use the managed endpoint. Webase will compile source files automatically as the agent writes them.

Add a custom connector from Settings → Connectors. Use https://www.webase.com/mcp/managed/messages and select OAuth as the auth method. After connecting, ask Claude things like “list my Webase apps” or “create a new Webase app for tracking gym workouts.”

ChatGPT (Custom GPT or Connector)

ChatGPT runs in a chat environment without local Node, so it should also use the managed endpoint. Configure the connector with the URLs from Authentication & Scopes; ChatGPT calls tools by name, so prompt naturally: “Use Webase to ship a landing page for an AI study buddy and deploy it.”

Codex (OpenAI CLI)

codex mcp add webase \
  --url https://www.webase.com/mcp/external/messages \
  --auth oauth

Codex is a desktop coding agent — same endpoint and reasoning as Claude Code.

Manual / scripts (x-api-token)

For shell scripts or internal automation, skip OAuth and use a personal API token (Production Admin → Users → Copy API Token):

curl -X POST https://www.webase.com/mcp/managed/messages \
  -H "Content-Type: application/json" \
  -H "x-api-token: $WEBASE_API_TOKEN" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/list"
  }'

What can the model do once connected?

• Read your roadmap and project memory (PM tools, all endpoints)
• Claim and complete features atomically (PM tools, all endpoints)
• Create a new Webase app — empty, no LLM-generated code
• Write source files (managed) or upload pre-compiled bundles (external)
• Define data models and CRUD records via MCP
• Run quality evaluations
• Deploy to Netlify and return the public URL

Tool selection happens by name and description, so you don't need to think about endpoints once configured — describe what you want and the host picks the right tool.