MCP Tools

When you connect your IDE to DYPAI via MCP, your AI assistant gains access to a specialized set of tools. These tools allow the AI to understand your project's structure and perform actions on your behalf.

Available Capabilities

You don't need to call these tools by name; simply describe what you want to do in natural language. The AI will select the appropriate tool to fulfill your request.

Database Management

Manage your PostgreSQL schema and data without writing SQL.

API & Endpoints

Build and manage your REST API layer directly from your editor.

Integrations & Credentials

Connect your backend to external services.

Storage & Files

Handle file uploads and bucket management.

System & History

Monitor and understand your project's activity.

How the AI uses these tools

The AI uses these tools as "building blocks". For a single request like "Create a full checkout system", the AI might:

1. Inspect your current database schema.
2. Create necessary tables (orders, line_items).
3. Define REST endpoints for the checkout process.
4. Configure a Stripe integration for payments.
5. Setup a WhatsApp notification flow for order confirmation.

Best Practices for Building Workflows with MCP

1) Node preflight

Before creating or updating endpoints:

• Run search_nodes to discover available node_type values.
• Use search_nodes with node_id to inspect node parameters and schemas.
• Never invent node_type. If a node is missing, choose a supported alternative.

2) Single placeholder contract

Use only these formats:

• ${input.field}
• ${vars.<variable>.field} (recommended when the node defines variable)
• ${nodes.<node_id>.field} (also supported for direct node references)
• ${current_user.field}
• ${current_user_id}

Avoid flat placeholders (${name}, ${email}, ${id}), as they cause validation or runtime failures.

Custom Query (SQL): In the dypai_database node with operation = custom_query, the query field uses ${variable} just like the rest of the workflow. Example: WHERE user_id = ${current_user_id} AND name ILIKE '%' || ${input.search} || '%'. Do not wrap placeholders in quotes: WHERE id = ${current_user_id} (correct), not WHERE id = '${current_user_id}' (incorrect). Available variables: ${input.field}, ${params.field}, ${current_user_id}, ${record_id}, ${limit}, ${offset}.

3) Validate before saving

Always validate workflows before persisting them. This catches:

• unknown node_type
• invalid operations
• placeholders outside the contract
• required-parameter inconsistencies

4) Credentials in integration nodes

For nodes that call external services (e.g., agent, slack, resend, whatsapp, telegram, google_sheets, stripe), include a valid credential_id in parameters.

Recommended pattern:

• credential_id = "${input.<provider>_credential_id}"
• Keep credentials in the project credentials vault; never hardcode API keys in workflow JSON.

5) Storage setup pattern

When a user asks to "add file storage" or "set up uploads", the correct flow is:

1. Create the bucket (if it doesn't exist) via the storage tool.
2. Create one endpoint named storage_{bucket} (e.g. storage_documents) with a single dypai_storage node:
   • operation = "${input.operation}" — routes upload, download, read, list, and delete automatically
   • bucket = "documents" — the bucket name
3. Tell the user to call dypai.storage.from('documents') from the SDK — the endpoint name and operation are sent automatically.

Do not create separate endpoints for upload, download, list, and delete. One endpoint per bucket handles everything.