Warning

The Simplicite MCP Server is a v7 feature. It is still in ALPHA, features will be added and improved This documentation is also in its early redaction stage.

MCP Server

An MCP Server is a lightweight bridge that exposes an application's capabilities as a set of callable tools, following the Model Context Protocol. In Simplicité's case, this means any MCP-compatible LLM can connect to a running Simplicité instance and interact with it autonomously: exploring a module's structure, reading and writing business objects, managing records, and even editing server-side Java code, all through natural-language requests instead of manual configuration or scripting.

Presentation

The Simplicité MCP Server exposes a set of tools to any compatible LLM, organized into three categories, enabling it to interact with a Simplicité application autonomously.

Discovery tools (get_ai_doc, get_md_documentation, find_object_by_name) let the LLM navigate a module: understand its structure, identify business objects, and orient itself before acting.

Manipulation tools cover standard CRUD operations as well as designer-oriented functions such as set_object_code, edit_object_code, and add_fields_to_object, which allow direct modification of a business object's configuration or code.

Finally, get_skill is the contextualization tool: it loads a targeted prompt into the LLM's context based on the nature of the request.

Tool	Category	Description
get_ai_doc	Discovery	Compact, LLM-optimized documentation for a module, object, or enum (fields, states, actions, hooks). Preferred over get_md_documentation.
get_md_documentation	Discovery	Human-readable Markdown description of a module. More verbose than get_ai_doc — use only when more detail is needed.
get_physical_mapping	Discovery	Physical database mapping of a module.
get_openapi	Discovery	Full OpenAPI 3.x spec for a module. Use only when API endpoint details are needed.
find_object_by_name	Discovery	Resolves a user-friendly name to the technical object code. Call first when the technical name is uncertain.
get_object_code	Designer	Retrieves the full Java source of a business object. Always call before editing.
set_object_code	Designer	Replaces the full Java source of a business object.
edit_object_code	Designer	Targeted search-and-replace on a business object's Java source. Preferred over set_object_code for partial edits.
add_fields_to_object	Designer	Adds fields to a business object (idempotent). Handles field, object-field link, and translations in a single call.
count_records	CRUD	Counts records for a business object with optional filters.
search_records	CRUD	Searches records and returns JSON with row_id values.
create_records	CRUD	Creates one or more records in a single call, with optional translations.
update_record	CRUD	Updates an existing record by row_id.
delete_record	CRUD	Deletes a record (irreversible).
call_action	CRUD	Invokes a business logic action on an object or record. Prefer over manual field updates when an action exists.
get_skill	Contextualization	Loads a targeted business prompt into the LLM's context based on the request type. Requires the MCP Server module.

Intended Workflow

To interact effectively with Simplicité, the LLM should follow this lifecycle:

1. Discovery:
   • Call get_ai_doc(name, type="module") to explore a module's objects, then get_ai_doc(name, type="object", sections=[...]) to get fields, states, actions, or hooks for a specific object. Fall back to get_md_documentation only if more detail is needed.
   • To create or edit Java logic, first call get_object_code to read the current source, then use edit_object_code for targeted changes or set_object_code for full rewrites.
2. Identification:
   • If the technical object name is uncertain, resolve it with find_object_by_name first.
   • Use search_records with filters to locate specific records and retrieve their row_id.
3. Operation:
   • Create objects with create_records (supports batch + translations in one call).
   • Add fields with add_fields_to_object (idempotent — safe to re-run).
   • Update or delete individual records with update_record / delete_record using the row_id from step 2.
   • Trigger business logic with call_action rather than manual field updates when an action exists for the operation.

Guiding the LLM towards best practices

In theory, an LLM could handle any request by freely combining discovery and manipulation tools. In practice, Simplicité often offers several ways to accomplish the same thing — and without guidance, the LLM may take incorrect paths or produce inconsistent, even non-functional results.

Skills solve this problem. Each skill is a targeted prompt that gives the LLM the right approach for a given type of request, avoiding the trial-and-error exploration phase.

To use get_skill, the MCP Server module must be installed from the AppStore (under Tools). This module exposes a McpPrompt object that stores skills. A starter dataset is provided, covering mainly maker needs (BUSINESS_OBJECT_CREATION, CROSSTAB, STATE_MODEL_CREATOR, etc.), along with a business skill example — CRM_CONTACTS, an agent that fills in demo contacts from a customer request written in natural language.

Improving over time

When an LLM handles a request, it either succeeds on the first try or requires several attempts. In the latter case, it is asked to write the skill itself for that type of request. The skill is then stored in McpPrompt and becomes immediately available for all similar requests going forward. The system therefore becomes more reliable the more it is used.

Built-in server configuration

The MCP endpoint is exposed as a servlet at:

YOUR_URL/mcp

warning

To expose it, the FEATURE_FLAG System Parameter needs to be updated:

{
	"mcp_server": true
}

For Visual Studio Code Copilot

Extensions --> MCP Servers --> Show Configuration (JSON)

Enter http://your_url/mcp (or https://your_instance.simplicite.io/mcp)

VSC should then open something_something/AppData/Roaming/Code/User/mcp.json

The json needs to be updated to handle authorization. Since OAuth2.1 is currently not supported by Simplicité, HTTP transport is bridged to STDIO using @nimbletools/mcp-http-bridge.

{
 "servers": {
     "SimpliciteServer": {
    			"description": "Access to a Simplicité instance with user rights",
    			"command": "npx",
				"args": [
				"-y",
				"@nimbletools/mcp-http-bridge",
				"--endpoint",
				"YOUR_ENDPOINT/mcp",
				"--token",
				"YOUR_TOKEN"
				]
  		}
 },
	"inputs": []
}

For Cursor

For Claude Desktop

Claude Desktop --> Settings --> Developer --> Edit Config --> claude_desktop_config.json

{
  "mcpServers": {
    "simplicite": {
      "description": "Access to a Simplicité instance with user rights",
        "command": "npx",
        "args": [
          "-y",
          "@nimbletools/mcp-http-bridge",
          "--endpoint",
          "YOUR_ENDPOINT/mcp",
          "--token",
          "YOUR_TOKEN"
          ]
      }
  }
}

For Gemini CLI

Gemini CLI supports HTTP transport natively. The server can be added using the following command:

gemini mcp add "Simplicité Java MCP Server" http://YOUR_URL/mcp --transport http --header "Authorization: Bearer YOUR_TOKEN"

Or by manually editing .gemini/settings.json:

{
  "mcpServers": {
    "Simplicité Java MCP Server": {
      "url": "http://YOUR_URL/mcp",
      "type": "http",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

warning

A Pro plan is required to add a custom remote connector.

For Mistral Vibe

warning

Only a remote instance can be added to Mistral.

warning

Organization admin rights are required to add a custom connector.

warning

The connector is shared through the organization with the given configuration. If EVERYONE in the Organization should not have access to the tools in the given instance with the given rights, DO NOT ADD IT.

Intelligence --> Connector --> Add Connector

Testing

Testing the tools relies on MCP Inspector.

In a terminal: npx @modelcontextprotocol/inspector

Select Streamable HTTP as Transport Type

Select YOUR_URL/mcp as URL

In Authentification, select Custom headers and enter:

{
  "Authorization": "Bearer YOUR_USER_TOKEN"
}