Issue Station
Sign In
Tools hereToolbox

claude-code - 💡(How to fix) Fix [BUG] Claude.ai MCP servers only partially loaded in v2.1.104 [1 comments, 2 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#47015•Fetched 2026-04-13 05:43:46
View on GitHub
Comments
1
Participants
2
Timeline
6
Reactions
0
Author
johnibjohnib
Participants
github-actions[bot]github-actions[bot]johnibjohnib
Timeline (top)
labeled ×3commented ×1mentioned ×1subscribed ×1

Error Message

Bug: With 10+ MCP servers configured in Claude.ai, only ~3 load per session. The rest are silently dropped — no error, no warning, no debug log entry. Root cause (per #35939): hardcoded batch concurrency of 3; servers not started before prompt processing begins are abandoned with no retry.

Root Cause

Bug: With 10+ MCP servers configured in Claude.ai, only ~3 load per session. The rest are silently dropped — no error, no warning, no debug log entry. Root cause (per #35939): hardcoded batch concurrency of 3; servers not started before prompt processing begins are abandoned with no retry.

Fix Action

Fix / Workaround

A pad-hack workaround exists (#29730) but there's no real fix. Please re-open or fix the underlying concurrency/batching logic.

RAW_BUFFERClick to expand / collapse

Still broken in v2.1.104 (latest). Tagging @bcherny since the bot-driven duplicate chain has buried this.

Duplicate chain — all locked, none fixed:

  • #40577 (my prior report) → closed as duplicate of #35939
  • #35939 → closed as duplicate of #29730
  • #29730 → closed for inactivity (stale) — the root issue was never actually fixed

Bug: With 10+ MCP servers configured in Claude.ai, only ~3 load per session. The rest are silently dropped — no error, no warning, no debug log entry. Root cause (per #35939): hardcoded batch concurrency of 3; servers not started before prompt processing begins are abandoned with no retry.

Steps to reproduce:

  1. Configure 10+ MCP servers in Claude.ai
  2. Launch Claude Code v2.1.104
  3. Observe only 3 servers loaded despite all being discovered in --debug log

Expected: all configured servers load Actual: only first batch of 3 loads, rest silently dropped

A pad-hack workaround exists (#29730) but there's no real fix. Please re-open or fix the underlying concurrency/batching logic.

Version: v2.1.104 | OS: macOS | Platform: Anthropic API

extent analysis

TL;DR

The most likely fix involves re-examining and adjusting the hardcoded batch concurrency limit to allow more than 3 MCP servers to load per session.

Guidance

  • Review the code related to batch concurrency in the Claude.ai configuration to understand how the limit of 3 is enforced and consider increasing this value to match the number of configured MCP servers.
  • Investigate the possibility of implementing a retry mechanism for servers that are not started before prompt processing begins, to prevent them from being abandoned silently.
  • Examine the --debug log to confirm that all MCP servers are indeed discovered but only the first 3 are loaded, as this will help in pinpointing the exact issue with the current concurrency logic.
  • Consider temporarily applying the pad-hack workaround mentioned in #29730 as a stopgap measure until a proper fix can be implemented.

Example

No specific code example can be provided without access to the Claude.ai source code, but the fix would likely involve modifying the batch concurrency limit and potentially adding a retry mechanism for late-starting servers.

Notes

The provided information suggests that the issue is well understood (hardcoded batch concurrency of 3), but a proper fix has not been implemented. The workaround from #29730 might help temporarily, but addressing the root cause is necessary for a reliable solution.

Recommendation

Apply the workaround from #29730 as a temporary measure, as there is no clear indication that a fixed version is available that addresses the concurrency issue directly. This workaround can help mitigate the problem until the underlying batching logic is revised.

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…
#api#environment variable#network issue#logging issue#authentication issue

Still need to ship something?

×6

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

Back to top recommendations

TRENDING