Claude Desktop

Claude can use the DevKit’s tools through MCP (Model Context Protocol), the standard way AI assistants use tools. You chat in plain English; Claude calls the tools; you confirm every request before anything is sent. Everything runs against the practice ABDM with made-up data.

Install

As an extension (easiest)

1. In the DevKit folder, build the extension:

   pnpm install && pnpm bundle:mcpb

2. Open build/abdm-devkit.mcpb with Claude Desktop, or go to Settings → Extensions → Install from file.

3. Confirm the install. The tools appear in Claude’s tools menu.

From a local build

1. Build the DevKit: pnpm install && pnpm build.

2. In Claude Desktop, open Settings → Developer → Edit Config and add the server, using full paths (which node shows yours):

   {
     "mcpServers": {
       "abdm-devkit": {
         "command": "/full/path/to/node",
         "args": ["/full/path/to/abdm-devkit/apps/mcp-server/dist/main.js"]
       }
     }
   }

3. Restart Claude Desktop.

Caution

Claude must be able to show you confirmation prompts (MCP calls this “elicitation”). If your version cannot, the tools that send requests stop with an error and nothing is sent.

Try these, one at a time

1. “What practice scenarios are available?”
2. “Request consent from asha.rao1@sbx for care management, for prescriptions and diagnostic reports from 2025, valid until the end of next month.” Confirm when asked.
3. “Check the status of that consent.” The simulated patient approves after a few seconds.
4. “Fetch the prescriptions under that consent.” Confirm when asked.
5. “Show me the records you received.”
6. “Request consent again, but this time the patient should refuse. Then try to fetch records under it.” The fetch is refused, as it should be.

The practice ABDM forgets everything when Claude Desktop restarts.