claude-code - 💡(How to fix) Fix [FEATURE] Remote Control: allow MCP server reload / process restart from a remote session [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#53125Fetched 2026-04-26 05:23:45
View on GitHub
Comments
0
Participants
1
Timeline
2
Reactions
0
Author
Participants
Timeline (top)
labeled ×2

When driving Claude Code from a Remote Control session, there is no way to refresh MCP server tool schemas or restart the Claude Code process — the operations that would do so (`/mcp`, killing + relaunching `claude`) all require physical access to the host machine.

For developers iterating on a custom MCP server (rebuild → exercise new tools → repeat), this forces a context switch to the host between every iteration, which defeats the value proposition of Remote Control.

Root Cause

When driving Claude Code from a Remote Control session, there is no way to refresh MCP server tool schemas or restart the Claude Code process — the operations that would do so (`/mcp`, killing + relaunching `claude`) all require physical access to the host machine.

For developers iterating on a custom MCP server (rebuild → exercise new tools → repeat), this forces a context switch to the host between every iteration, which defeats the value proposition of Remote Control.

RAW_BUFFERClick to expand / collapse

Summary

When driving Claude Code from a Remote Control session, there is no way to refresh MCP server tool schemas or restart the Claude Code process — the operations that would do so (`/mcp`, killing + relaunching `claude`) all require physical access to the host machine.

For developers iterating on a custom MCP server (rebuild → exercise new tools → repeat), this forces a context switch to the host between every iteration, which defeats the value proposition of Remote Control.

Reproduction

  1. Start `claude` in remote-control mode on host A.
  2. Connect from device B (different machine, browser / mobile).
  3. Rebuild a custom MCP server's binary on host A while the session is live.
  4. From device B, attempt to refresh MCP tool schemas — there is no remote-eligible path:
    • `/mcp` is local-only (documented).
    • There is no `/mcp reload` or `/reload-mcp` command.
    • There is no remote command to restart the Claude Code process; even if one existed, the remote session would need to auto-reconnect after the process respawned.
  5. The only way forward is to physically return to host A, kill + relaunch `claude`, and reconnect from device B.

Proposed

Any one (or combination) of the following would close the gap, ranked by my preference:

  1. Make MCP reload remote-eligible. A non-interactive variant of `/mcp` (e.g. `/mcp reload`, or a flag-driven form) that re-handshakes the MCP children without an interactive picker. This would also benefit related issues:
    • #46426 — Feature request: hot-reload MCP servers without restarting the session
    • #40059 — Allow reloading MCP servers without restarting the session
    • #36847 — SIGHUP handler for config and plugin reload (headless-friendly)
  2. Remote process restart with auto-resume. A remote command that triggers `claude` to re-exec itself (or a wrapper that respawns it), with the Remote Control session auto-reconnecting to the new process. Mirrors the precedent of #39724 (`/model` switching in remote-control).
  3. HTTP / WS endpoint on the local Claude Code process that the Remote Control bridge can call to issue an MCP reload command on the user's behalf, gated by the same auth as the Remote Control session itself.

Context / why now

Custom MCP servers — including community projects like vaultpilot-mcp and many others driving the recent growth in MCP usage — ship fresh schemas with every PR. The "remote drives the host" pattern is exactly what Remote Control sells, but the iteration loop today is "remote → can't reload → walk to host → restart → walk back → continue", which is meaningfully worse than just developing locally.

Related

  • #46426 — Feature request: hot-reload MCP servers without restarting the session (general; not remote-specific)
  • #40059 — Allow reloading MCP servers without restarting the session (general)
  • #36847 — feat: SIGHUP handler for config and plugin reload (headless-friendly alternative)
  • #39724 — [FEATURE] Allow /model switching in remote-control sessions (precedent: expand a local-only command to remote-eligible)
  • #45989 — Remote-control: web client never delivers work to bridge after service restart (related infra: how Remote Control survives process churn today)

If reload + auto-resume already exists and I missed the doc page, happy to close — but the answer I got from the in-product help was "you have to restart locally; `/mcp` is local-only."

extent analysis

TL;DR

Implement a remote-eligible command to reload MCP server schemas or restart the Claude Code process without requiring physical access to the host machine.

Guidance

  • Consider adding a non-interactive variant of the /mcp command, such as /mcp reload, to allow remote reloading of MCP servers.
  • Explore implementing a remote process restart with auto-resume, enabling the Remote Control session to reconnect to the new process after restart.
  • Evaluate creating an HTTP/WS endpoint on the local Claude Code process for the Remote Control bridge to issue an MCP reload command.

Example

No code snippet is provided as the issue does not contain specific code references.

Notes

The proposed solutions aim to address the limitation of requiring physical access to the host machine for reloading MCP server schemas or restarting the Claude Code process. The feasibility of these solutions may depend on the existing architecture and implementation of the Remote Control feature.

Recommendation

Apply a workaround by implementing a remote-eligible command to reload MCP server schemas, as this seems to be the most straightforward solution to address the current limitation.

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