Two claude.ai connectors of the same service type (Notion) collide — only one loads per session

Status Open
Maintainer reply None cached
Activity 1 comment · opened Jul 25, 2026

Setup: Two separate Notion workspace connectors authorized under Settings > Connectors in claude.ai, both showing as connected (green check) in the connector settings UI:

  • "Notion"
  • "Notion (Everything Workspace)"

Expected: Both connectors load as separate MCP servers in a Claude Code session (interactive or headless claude -p), giving access to both Notion workspaces simultaneously.

Actual: claude mcp list only ever shows ONE of the two ("Notion (Everything Workspace)"). The other ("Notion") does not appear as failed or "needs authentication" — it's simply absent from the list, despite being connected at the account/connector-settings level. This is reproducible across both interactive sessions and headless claude -p --dangerously-skip-permissions cron invocations.

Ruled out:

  • Not an auth/token issue — reconnecting/reauthorizing the missing connector in claude.ai settings does not change the behavior; it still shows connected there but absent from claude mcp list.
  • Not ANTHROPIC_API_KEY/ANTHROPIC_AUTH_TOKEN auth-precedence — neither is set in the environment, shell profile, or launchd job plist for the headless cron job affected.
  • Not a stale local ~/.claude.json config issue — removing duplicate manually-added mcpServers entries for the same URLs did not surface the missing connector.

Suspected cause: Claude Code appears to dedupe or otherwise only load one connector per underlying service type (e.g., "Notion") rather than treating each authorized connector instance independently, even though claude.ai's own connector settings clearly support and display multiple distinct connections of the same type.

Impact: Any user who legitimately needs two workspaces of the same connector type available in one session (e.g., a business workspace + a personal workspace) cannot do so — one connector is silently dropped every session with no error surfaced to explain why.

Workaround in use: bypassing the claude.ai OAuth connector entirely for one of the two workspaces, in favor of a locally-configured MCP server (@notionhq/notion-mcp-server) authenticated via a Notion internal integration token instead of OAuth, added directly to ~/.claude.json's mcpServers. This avoids the collision since it isn't a "claude.ai connector" at all, but it shouldn't be necessary — the underlying claude.ai connector limitation is the real bug.

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗