claude-code - 💡(How to fix) Fix [BUG] MCP tools (Notion, Slack, and other configured MCP servers) are unreliable in Claude Code cloud env sessions [1 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
anthropics/claude-code#46203Fetched 2026-04-11 06:26:26
View on GitHub
Comments
0
Participants
1
Timeline
4
Reactions
0
Author
Participants
Timeline (top)
labeled ×4

Error Message

Error Messages/Logs

No explicit error messages are surfaced to the user or assistant. The failures are silent:

  • No error, no "connecting..." indicator — just empty results

Code Example

No explicit error messages are surfaced to the user or assistant. The failures are silent:

1. DELAYED REGISTRATION (session start):
   - Assistant calls: ToolSearch(query="notion")
   - Response: "No matching deferred tools found"
   - No error, no "connecting..." indicator — just empty results
   - Tools appear silently in a later system-reminder

2. REPEATED DISCONNECTIONS (mid-session):
   - System-reminder arrives: "The following deferred tools are no longer available 
     (their MCP server disconnected). Do not search for them — ToolSearch will 
     return no match: mcp__fb5f90cd-...__notion-search, [60+ tools listed]"
   - Minutes later, another system-reminder: "The following deferred tools are now 
     available via ToolSearch."
   - This cycle repeated 3+ times in a single ~15 minute conversation.
   - All 3 MCP servers (Notion fb5f90cd-*, Slack da116102-*, and 66450a78-*) 
     always disconnect and reconnect together.

3. SUBAGENT ABSENCE:
   - Subagent runs ToolSearch(query="notion") — returns no results
   - Subagent's only deferred tools: AskUserQuestion, ExitPlanMode, Monitor, 
     NotebookEdit, RemoteTrigger, TodoWrite, WebFetch, WebSearch
   - No MCP tools are ever registered for the subagent session
   - Subagent tries WebFetch on Notion URL — gets HTTP 403 (expected, 
     since it's private/authenticated)
RAW_BUFFERClick to expand / collapse

Preflight Checklist

  • I have searched existing issues and this hasn't been reported yet
  • This is a single bug report (please file separate reports for different bugs)
  • I am using the latest version of Claude Code

What's Wrong?

MCP tools (Notion, Slack, and other configured MCP servers) are unreliable in Claude Code Cloud Environment sessions:

  1. Delayed registration — At session start, ToolSearch returns no results for MCP tools. They appear in a later system-reminder after the MCP server finishes connecting, but the assistant has already told the user that no MCP tools exist.
  2. Repeated disconnections — All MCP servers disconnect and reconnect multiple times mid-conversation, making them unusable during gaps.
  3. Complete absence in subagent sessions — Subagents spawned via the Agent tool never receive MCP tools. Only base deferred tools (AskUserQuestion, TodoWrite, etc.) are registered for them.

What Should Happen?

  • MCP tools should be discoverable via ToolSearch from the first user message, or the assistant should be informed that MCP servers are still connecting.
  • MCP servers should maintain stable connections throughout a session.
  • Subagent sessions should inherit MCP tool access from the parent session.

Error Messages/Logs

No explicit error messages are surfaced to the user or assistant. The failures are silent:

1. DELAYED REGISTRATION (session start):
   - Assistant calls: ToolSearch(query="notion")
   - Response: "No matching deferred tools found"
   - No error, no "connecting..." indicator — just empty results
   - Tools appear silently in a later system-reminder

2. REPEATED DISCONNECTIONS (mid-session):
   - System-reminder arrives: "The following deferred tools are no longer available 
     (their MCP server disconnected). Do not search for them — ToolSearch will 
     return no match: mcp__fb5f90cd-...__notion-search, [60+ tools listed]"
   - Minutes later, another system-reminder: "The following deferred tools are now 
     available via ToolSearch."
   - This cycle repeated 3+ times in a single ~15 minute conversation.
   - All 3 MCP servers (Notion fb5f90cd-*, Slack da116102-*, and 66450a78-*) 
     always disconnect and reconnect together.

3. SUBAGENT ABSENCE:
   - Subagent runs ToolSearch(query="notion") — returns no results
   - Subagent's only deferred tools: AskUserQuestion, ExitPlanMode, Monitor, 
     NotebookEdit, RemoteTrigger, TodoWrite, WebFetch, WebSearch
   - No MCP tools are ever registered for the subagent session
   - Subagent tries WebFetch on Notion URL — gets HTTP 403 (expected, 
     since it's private/authenticated)

Steps to Reproduce

  1. Configure Notion MCP (and optionally Slack + other MCP servers) in Claude Code web
  2. Start a new session
  3. Immediately ask: "Use the Notion MCP to search for X"
  4. The assistant runs ToolSearch with query "notion" — returns no results
  5. The assistant tells you no Notion MCP is configured
  6. Moments later, a system-reminder appears listing all MCP tools as available
  7. Additionally: during the session, all MCP servers disconnect and reconnect multiple times (observed 2+ disconnect/reconnect cycles in a single conversation)
  8. Additionally: spawn a subagent via the Agent tool and ask it to use MCP tools — it will never find them

Claude Model

Not sure / Multiple models

Is this a regression?

I don't know

Last Working Version

Not Sure

Claude Code Version

Claude 1.1617.0 (8d6345) 2026-04-09T16:10:15.000Z

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

  • All three configured MCP servers (Notion fb5f90cd-, Slack da116102-, and 66450a78-*) connect and disconnect in unison, suggesting a shared transport/lifecycle issue rather than an individual server problem.
  • The subagent issue is particularly impactful since it makes it impossible to delegate MCP-dependent work to background agents.

Claude-code itself has compiled this information for me, its hard to tell if what it claims is correct.

extent analysis

TL;DR

The most likely fix involves addressing the shared transport/lifecycle issue affecting all MCP servers, ensuring they maintain stable connections and are discoverable via ToolSearch from the start of a session.

Guidance

  1. Investigate MCP server connection handling: Review the code responsible for connecting and disconnecting MCP servers to identify potential issues causing repeated disconnections and delayed registrations.
  2. Verify ToolSearch implementation: Examine the ToolSearch function to ensure it correctly handles cases where MCP servers are still connecting or have temporarily disconnected.
  3. Subagent session configuration: Check the configuration and initialization process for subagent sessions to understand why they do not inherit MCP tool access from the parent session.
  4. Review system-reminder logic: Analyze the system-reminder mechanism to prevent silent failures and ensure that users and assistants are informed about MCP server connection status.
  5. Test MCP server connections independently: Isolate and test each MCP server's connection to identify if the issue lies with a specific server or the shared transport/lifecycle mechanism.

Example

No specific code snippet can be provided without more context, but ensuring that the ToolSearch function can handle asynchronous MCP server connections might involve using callbacks or promises to wait for the connection to establish before returning results.

Notes

The issue's complexity and the lack of explicit error messages make it challenging to provide a precise solution. The guidance points are aimed at helping to identify and potentially resolve the underlying causes.

Recommendation

Apply a workaround to improve the handling of MCP server connections and ToolSearch functionality, as the root cause of the issue is not immediately clear and may require further investigation.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING

claude-code - 💡(How to fix) Fix [BUG] MCP tools (Notion, Slack, and other configured MCP servers) are unreliable in Claude Code cloud env sessions [1 participants]