Custom MCP connector fails to authenticate despite server confirmed fully functional via manual protocol test

### MCP server URL or name

https://example.com/wp-json/fluent-crm/mcp *(actual domain redacted for this report)

### How are you connecting to this server?

From a custom connector URL I entered manually

### Where does the issue occur?

When pressing the Connect button

### When did you last reproduce this?

Mon Sep 7 10:36:55 CDT 2026

### Browser and OS

Chrome 152, macOS Tahoe version 26.6.2

### Describe the issue

**Summary:**
Claude's custom connector ("Add custom connector" in Settings → Connectors) fails to connect to a third-party MCP server under both supported authentication paths, even though the server has been independently verified to work correctly with the exact same credentials via a manual multi-step MCP session test.

**Server details:**
- Server: FluentCRM MCP Server (WordPress plugin, WPManageNinja)
- Endpoint: https://example.com/wp-json/fluent-crm/mcp *(actual domain redacted for this report)*
- Transport: Streamable HTTP

**What I tried:**

1. **OAuth ("Always required" — Detected):** Both "Use Anthropic's hosted client metadata" and "No client ID — register one automatically" (DCR) failed with "Couldn't register with [server]'s sign-in service." Expected, since the server vendor confirmed no OAuth authorization server is implemented for this endpoint.

2. **"None" authentication + manual "Authorization: Basic " header:** Described in the connector UI as the path "for servers that use an API key instead of OAuth." Filled in correctly, saved, connector removed and re-added fresh multiple times, app fully restarted between attempts — consistently fails with: "Couldn't connect to the server. Check that the URL points to a valid MCP server." Reference IDs: ofid_a25800e3625dd0ab, ofid_948f519483e0ed8c

**Proof the server and credentials work correctly:**

I manually replicated the full MCP session lifecycle via curl using the identical Basic Auth credential:

- `initialize` → HTTP 200, valid JSON-RPC response, server returns an `mcp-session-id` header
- `notifications/initialized` (same session ID) → HTTP 202 Accepted
- `tools/list` (same session ID) → HTTP 200 with a complete, well-formed tool list

This rules out invalid credentials, missing permissions, a malformed endpoint, or a broken server-side MCP implementation — independently confirmed by the server vendor as well.

**Notable comparison:** The identical "None + manual Basic Auth header" pattern worked successfully for a different MCP server (Fluent Support, also WPManageNinja) in this same app, after an initial server-side permissions issue was resolved. So this auth mode isn't universally broken — something specific to this connector/server combination fails during the app's actual connection attempt, despite the server responding correctly to the same request pattern outside the app.

**Ask:** Could someone investigate what's failing between the app's connection attempt and the server, given the server side is independently confirmed correct? Happy to provide further logs or test additional scenarios.

### Error details

```shell
Couldn't reach Fluent CRM. You can check the server URL and verify the server is running. If this persists, share this reference with support: "ofid_c8c7e0f7382f83fa"
```

### Expected behavior

Clicking "Connect" with a valid custom header (Authorization: Basic ) under "None" authentication should establish a working MCP connection to the server, the same way it does for other MCP servers configured this way (e.g., Fluent Support, which connects successfully in this same app using the identical auth pattern). Since the server has been independently verified — via a full manual initialize → notifications/initialized → tools/list session over curl using the exact same credential — to respond correctly at every step of the MCP protocol, I expected the in-app connector to succeed as well, rather than fail immediately with a generic "couldn't reach server" error.

### Additional context