fix(mcp): register call_tool as the handler (every MCP tool call was crashing)

[sbhooley]

Fixes the misplaced @server.call_tool() decorator that made every MCP tool call crash.

Problem

All memory_*, ainl_*, and a2a_* tools fail with:

_ainl_unavailable_payload() takes 0 positional arguments but 2 were given

In mcp_server/server.py the @server.call_tool() decorator was attached to the 0-arg helper _ainl_unavailable_payload() instead of the async dispatcher call_tool(name, arguments). The SDK registered the helper as the handler and invoked it with (name, arguments) → TypeError on every call. Hooks and direct DB writes were unaffected (they bypass the dispatcher), which masks the breakage.

Fix

Move the decorator onto call_tool. _ainl_unavailable_payload remains a plain helper, still called with 0 args from the ainl_* fallback branches.

Verified

• py_compile clean.
• Imported the patched module in a separate process and called call_tool("memory_store_semantic", {...}) → returned a node_id (no TypeError).
• Confirmed live in Claude Code after killing the stale mcp_server.server process and reloading.

Note for applying the fix

/reload-plugins does not restart an already-running MCP server subprocess — the existing mcp_server.server process must be killed (or Claude Code fully restarted) for the patch to load.

Closes #1

🤖 Generated with Claude Code