MCP Model Explorer¶

The Model Explorer is the HTTP/SSE MCP server that lets AI assistants discover, inspect, and operate on your AMSDAL models via natural language.

Available Tools¶

list_available_models¶

Discover all registered AMSDAL models with their fields and relationships.

Returns a list of models with: - Model name - Field names - Foreign key relationships

get_model_schema¶

Get the full JSON Schema for a specific model, including computed properties.

Parameters:

Parameter	Type	Description
`model_name`	`str`	Name of the AMSDAL model

Returns the complete JSON Schema with field types, constraints, defaults, and computed properties.

get_model_relationships¶

Inspect a model's foreign key graph — both outgoing (many-to-one) and incoming (one-to-many) relationships.

Parameters:

Parameter	Type	Description
`model_name`	`str`	Name of the AMSDAL model

Returns outgoing and incoming relationship details with field names, target models, and cardinality.

get_model_field_values_by_ids¶

Fetch specific field values for objects by their IDs. Supports history.

Parameters:

Parameter	Type	Description
`model_name`	`str`	Name of the AMSDAL model
`field_name`	`str`	Field name to fetch values for
`object_ids`	`list[str]`	Object IDs to fetch

perform_crud_operation¶

Execute CRUD operations using natural language with a two-step confirmation flow.

Parameters:

Parameter	Type	Description
`model_name`	`str`	Name of the AMSDAL model
`operation`	`str`	One of: `read`, `create`, `update`, `delete`
`query`	`str`	Natural language description of the operation
`confirm`	`bool`	Set to `true` to execute after preview (default `false`)
`preview_limit`	`int`	Max records to show in preview (default `5`)
`limit`	`int \\| None`	Max records to return (default `30`)
`analysis`	`dict \\| None`	Analysis from step 1, passed back in step 2 (default `None`)

Two-step flow:

First call with confirm=false — returns a preview with an analysis dict
Second call with confirm=true and the analysis from step 1 — executes the operation

This prevents accidental data modifications through AI assistants.

Example Conversation¶

A typical conversation with an AI assistant connected via MCP:

User: What models are available?

Assistant calls list_available_models

Assistant: You have Customer, Order, Product, and Invoice models.

User: Show me all customers from New York

Assistant calls perform_crud_operation(model_name='Customer', operation='read', query='customers from New York', confirm=false)

Assistant: Found 12 customers from New York. [shows preview]

User: Update John Doe's email to john@new.com

Assistant calls perform_crud_operation(model_name='Customer', operation='update', query='update John Doe email to john@new.com', confirm=false)

Assistant: This will change John Doe's email from john@old.com to john@new.com. Confirm?

User: Yes

Assistant calls with confirm=true

Row-Level Filtering¶

The Model Explorer uses api_objects manager, which applies row-level permissions automatically. Users only see and modify records they're authorized to access.