Claude Code plugin writes memories but recall/search returns empty — API key and OAuth both affected

[FoodGoods]

Description

Claude Code plugin successfully writes documents (3,060+ docs, 91+ memories visible in Nova UI at app.supermemory.ai), but all read paths return empty:

• API search (v3 and v4, with and without containerTag, threshold: 0.0) returns zero results
• MCP recall tool via OAuth returns "No memories or profile found"
• Direct API queries with searchMode: "hybrid" return nothing

Write path works perfectly. Read path is completely broken.

Likely Related

This appears to be the same root cause as #738 (API key integration creates isolated agent space). That issue is closed (ENG-576) but the problem persists for early adopters who signed up before the consumer/developer org unification.

Steps to Reproduce

1. Sign up early (pre-org-unification), set up API key auth (SUPERMEMORY_CC_API_KEY)
2. Install claude-supermemory plugin — plugin captures tool usage and conversations
3. Verify docs/memories appear in Nova UI at app.supermemory.ai — they do
4. Install MCP server via npx install-mcp@latest --client claude-code --oauth=yes
5. Attempt recall via MCP — returns "No memories or profile found"
6. Attempt direct API search (v3/v4) with the API key — returns zero results
7. Attempt "Migrate from MCP v1" in Connections & MCP settings — fails with "Failed to migrate documents"

Migration UX Feedback

The "Migrate from v1" button is the only path to fix this for early adopters, but:

• It's buried in Connections & MCP → Connect your AI → Migrate from v1 — not discoverable
• The "Learn More" button next to it links to the generic MCP overview page, not migration documentation
• There are no docs explaining the v1 → v2 migration process, what the old MCP link format is, or how to find your v1 user ID
• The migration dialog asks for a URL (https://mcp.supermemory.ai/userId/sse) but doesn't tell you where to find your userId
• When migration fails, the error ("Failed to migrate documents") gives no detail about why

For early adopters, this should be a prominent, well-documented path — ideally a single button on the main dashboard that detects v1 data exists and offers to migrate it automatically.

Environment

• claude-supermemory v0.0.2, macOS
• Pro plan, early adopter (pre-org-unification)
• mcp-remote v0.1.37 for OAuth
• MCP server shows connected, whoAmI returns valid user

Expected Behavior

Memories written by the plugin should be retrievable via the API and MCP recall tool. Early adopters should have a clear, documented migration path that works.

[linear]

ENG-672