[BUG] Atlassian MCP connector: write calls (addCommentToJiraIssue) blocked by Cloudflare; connector also intermittently drops to "no OAuth token" during reconnect

Status Open
Reported on v2.1.217
Maintainer reply None cached
Activity 1 comment · opened Jul 22, 2026

Preflight Checklist:

  • [x] I have searched existing issues and this hasn't been reported yet (related but distinct: #79993, which describes a different symptom — see note below)
  • [x] This is a single bug report
  • [x] I am using the latest version of Claude Code

What's Wrong?

The claude.ai Atlassian MCP connector (Jira, proxied through mcp-proxy.anthropic.com) works for read calls but reliably fails on write calls. addCommentToJiraIssue fails nearly every time with a Cloudflare "Sorry, you have been blocked" challenge page returned as the HTTP response body, instead of a normal API response or error.

Reproducible:

  • Across two separate days
  • With different comment content each time (rules out a content/payload trigger)
  • Immediately after a full connector reconnect (/mcp disconnect + reconnect)
  • Immediately after a full re-authentication (/mcp disconnect + fresh OAuth login)

Reconnecting and re-authenticating do not resolve it, suggesting the block happens upstream of the connector's auth state.

A secondary, possibly related issue: after one Cloudflare-blocked write call, the SSE connection dropped and the automatic reconnect loop churned through several attempts returning a different error — an auth/token error — before eventually reconnecting successfully:

claude.ai proxy connection failed: Streamable HTTP error: Error POSTing to endpoint: {"type":"error","error":{"type":"authentication_error","message":"MCP server requires authentication but no OAuth token is configured.","details":{"error_code":"mcp_unauthorized_no_token"}}}

This repeated across 5 reconnection attempts (exponential backoff: 1s, 2s, 4s, 8s) before succeeding. The very next write call after the successful reconnect hit the same Cloudflare block again.

What Should Happen?

Write calls to Jira via this connector should succeed reliably, the same as read calls do. If Cloudflare is intentionally rate-limiting or challenging this traffic pattern, the connector should handle the challenge transparently rather than surfacing raw HTML to the tool caller. The mcp_unauthorized_no_token error also shouldn't occur when the connector was working moments before with a valid, active session.

Error Messages/Logs

From ~/Library/Caches/claude-cli-nodejs/<project>/mcp-logs-claude-ai-Atlassian/*.jsonl:

Read calls succeed normally:

Calling MCP tool: getJiraIssue
Tool 'getJiraIssue' completed successfully in 2s

Write calls fail with a Cloudflare block (Ray ID differs each attempt, confirming each request reaches Cloudflare fresh rather than hitting a cached failure):

Calling MCP tool: addCommentToJiraIssue
CLAUDEAI-PROXY connection dropped
Connection error: Streamable HTTP error: Error POSTing to endpoint: <!DOCTYPE html>...
<h1>Sorry, you have been blocked</h1>
<h2>You are unable to access anthropic.com</h2>
...
Cloudflare Ray ID: [redacted — multiple distinct Ray IDs observed across attempts]

Notably, the Cloudflare block page's own heading says "You are unable to access anthropic.com" — not atlassian.net — suggesting this is happening on Anthropic's own mcp-proxy.anthropic.com infrastructure (confirmed independently: that hostname resolves and responds with server: cloudflare headers), not on Jira's side or any client-side network/proxy.

Steps to Reproduce

  1. Connect the claude.ai Atlassian connector (Jira) via /mcp.
  2. In a Claude Code session, call a Jira read tool (e.g. getJiraIssue) — succeeds normally.
  3. Call a Jira write tool (e.g. addCommentToJiraIssue) on the same issue — fails with a Cloudflare "Sorry, you have been blocked" HTML page as the tool error, instead of a normal API response.
  4. Retry the same write call — fails again with a new Cloudflare Ray ID.
  5. Run /mcp to disconnect and reconnect (or fully re-authenticate) — read calls continue to work, but the next write call still fails the same way.

Is this a regression? I don't know

Claude Code Version: 2.1.217 (Claude Code)

Platform: Anthropic API

Operating System: macOS

Additional Information

This looks distinct from #79993, which describes the connector never listing any Atlassian tools at all (clean handshake, hasTools: true, then connection closes with no error, zero tools ever exposed). In the case reported here, tools list correctly and reads succeed — only write calls are blocked, with an actual Cloudflare challenge response rather than a silent tool-discovery failure.

View original on GitHub ↗

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