AI agent tools for AL development

Applies to: AL Language extension 17.0 and later

AI agent tools enable GitHub Copilot and any Model Context Protocol (MCP)-compatible AI agent to perform core Business Central development tasks on your behalf—building, publishing, searching symbols, running diagnostics, and debugging AL extensions—without leaving your editor or workflow.

These tools are designed for three usage surfaces:

The MCP and Visual Studio Code tools share a consistent set of tool names, so an agent trained on one surface can work with the other with no retraining. The AL LSP Server complements them by providing deep semantic understanding of the codebase rather than discrete task execution. Learn more in AL LSP.

Available tools

The following tools are available across both surfaces. Some tools are exclusive to one surface due to the nature of the operation.

Using tools in Visual Studio Code with GitHub Copilot

When GitHub Copilot is running in Agent mode, it can call AL tools automatically in response to natural-language requests. You can also reference a tool explicitly in Copilot Chat by prefixing its name with #:

#al_build
#al_symbolsearch
#al_getdiagnostics

Example Copilot prompts:

• "Build my AL project and show me any errors." — Copilot calls al_build, then al_getdiagnostics.
• "Find all codeunits related to posting." — Copilot calls al_symbolsearch with appropriate filters.
• "Publish my extension to my sandbox environment." — Copilot calls al_publish.

Copilot can chain tools automatically. For example, if al_build reports errors, Copilot might call al_getdiagnostics to retrieve the full details, suggest code fixes, rebuild, and then publish—all in a single conversation turn.

Using the AL MCP Server

The AL MCP Server is a standalone process that exposes AL tools over the Model Context Protocol. It runs anywhere that .NET 8 is available and doesn't require Visual Studio Code.

Starting the server

altool launchmcpserver --transport stdio

JSON-RPC traffic is exchanged on stdout. All diagnostic logs are written to stderr. The server shuts down cleanly on EOF or Ctrl+C.

Connecting an agent

Configure your MCP-compatible agent (for example, Claude Desktop, a custom agent built on the MCP SDK, or a CI pipeline agent) to launch the command above as an MCP server. The agent discovers the available AL tools automatically through the MCP tools/list call.

Authentication

When the AL MCP Server needs to connect to a Business Central cloud environment, it uses browser-based interactive authentication via Microsoft Authentication Library (MSAL). Tokens are cached securely on disk and reused across tool calls.

• Call al_auth_login to authenticate before calling al_publish or al_downloadsymbols with a cloud environment.
• For on-premises deployments using Windows authentication, no explicit sign in is required.
• Call al_auth_logout to clear cached tokens when switching accounts.

Common workflows

Build and validate

1. Call al_downloadsymbols to download symbol packages if they're missing or out of date.
2. Call al_build (or al_compile in AL MCP for faster validation without producing an .app).
3. Call al_getdiagnostics to inspect errors and warnings.

Build and publish to a sandbox

1. Call al_downloadsymbols to refresh symbols.
2. Call al_build to generate the .app package.
3. Call al_auth_login (AL MCP only) to authenticate to the cloud environment.
4. Call al_publish with the target environment details.

Symbol discovery

1. Call al_symbolsearch with a query to find tables, codeunits, pages, fields, or other AL objects across the project and its dependencies.

Debug an already-published extension (Visual Studio Code)

1. Call al_debug to attach the debugger to the deployed extension without republishing.
2. Optionally call al_setbreakpoint to set breakpoints at specific lines.

Security

Authentication tokens are never passed as tool parameters or returned in tool responses. When a tool requires access to a Business Central cloud environment, it triggers a browser-based Microsoft Entra ID sign-in flow. The resulting tokens are stored by MSAL in a secure local cache and reused transparently.

To force re-authentication, set noCache: true on any tool that supports it, or call al_auth_logout first.

AL LSP for semantic code intelligence

In addition to the task-oriented tools listed above, ALTool can launch a full Language Server Protocol (LSP) server via the launchlspserver command. While the MCP tools perform discrete actions (build, publish, search), the LSP server provides continuous semantic code intelligence—hover, go-to-definition, find-references, completions, rename, and type hierarchy—over JSON-RPC on stdio.

The AL LSP server is the best choice when an agent needs to navigate and understand AL code structurally rather than perform predefined tasks. It correctly resolves cross-project relationships such as internalsVisibleTo and propagateDependencies, making it more reliable than text-based search for symbol resolution in multi-project workspaces.

Learn more in AL LSP.

Related information

AL Language Model Tools for Visual Studio Code
AL MCP Server reference
AL tool reference: al_build
AL tool reference: al_compile
AL tool reference: al_publish
AL tool reference: al_downloadsymbols
AL tool reference: al_symbolsearch
AL tool reference: al_getdiagnostics