Skip to content

Application tools and AI development helpers

Aksara's stable synchronous MCP tools and its experimental AI development helpers are different surfaces. Do not assume a tool-shaped JSON example is an executable protocol contract.

Generated application tools

Selected ViewSets expose generated CRUD and custom-action tools through the MCP protocol server. The actual names and input schemas come from client discovery. For example, the ticket-desk client discovers names such as ticket_create; there is no universal create_record MCP command that accepts an arbitrary model name.

Use /mcp/ with the official client. /ai/tools/mcp is a separate HTTP JSON inspection catalog. Discovery visibility does not replace execution-time identity, permission, policy, tenant and field enforcement. Your application must verify credentials and resolve the server-owned Principal.

The ticket-desk chapter executes real tool calls and tests denied writes, expired credentials, changed permissions and tenant boundaries. It needs no LLM provider. Synchronous MCP does not automatically turn these calls into Durable Operations or protocol-level MCP Tasks.

Experimental development helpers

Context export, schema analysis, prompt packs, plans and code patches belong to the experimental development surfaces. Their data structures are not a generic MCP command catalog. In particular, names such as generate_test, get_settings or run_migration in a conceptual plan do not imply an exposed application tool.

Use the specific guide for the intended operation:

Do not expose development or migration capabilities to an agent merely because it can describe them in a prompt. An application-owned custom tool needs a reviewed execution and authorization path just like any other endpoint.