rdmsm4x-changelog-20260831-2204-remove-broken-stitch-mcp-fleet-wide

Google Stitch MCP and associated schemas/skills were causing tool execution failures and breaking agent workflows across fleet harnesses; completely removed Stitch MCP registrations, skill files, and active processes across all 6 fleet hosts.

Scope

• Fleet hosts: rdmsm4x, rdmbair15m5, rdmbair13m5, jdmbair13m5, rdmpw3265m, rdmpw3275m
• Harnesses cleaned: Claude Code (~/.claude.json), Claude Desktop (claude_desktop_config.json), OpenAI Codex (~/.codex/config.toml), Antigravity / Gemini CLI & IDE (~/.gemini/config/mcp_config.json, ~/.gemini/antigravity*/mcp/)

Exact Files Touched

• Configuration files purged of stitch / StitchMCP:
  • ~/.claude.json (server registration and disabledMcpServers lists across projects)
  • ~/Library/Application Support/Claude/claude_desktop_config.json
  • ~/.codex/config.toml ([mcp_servers.stitch] and [mcp_servers.stitch.env] tables removed)
  • ~/.gemini/config/mcp_config.json
• Directories deleted:
  • ~/.gemini/antigravity-cli/mcp/stitch
  • ~/.gemini/antigravity/mcp/stitch
  • ~/.gemini/config/skills/google-stitch
  • ~/.gemini/config/plugins/user-skills/skills/google-stitch
  • ~/.gemini/skills/google-stitch
  • ~/.claude/skills/google-stitch
  • ~/.codex/skills/google-stitch
• Scripts updated:
  • ~/dev/fleet/maintenance/scripts/remove_stitch_mcp.py (created and deployed to all hosts)
  • ~/dev/fleet/maintenance/scripts/apply_mcp_disable_v1.0.py (removed stitch from targets)
  • ~/dev/fleet/tooling/host-scripts/*/apply_mcp_disable_v1.0.py

Commands Run & Actions Taken

1. Created automated host-local removal tool remove_stitch_mcp.py with atomic JSON/TOML modification and pre-mutation backups.
2. Executed removal across all 6 machines: rdmsm4x, rdmbair15m5, rdmbair13m5, jdmbair13m5, rdmpw3265m, rdmpw3275m.
3. Terminated orphan stitch-mcp background node processes on rdmbair15m5 (PIDs 88786, 88655) and rdmpw3275m (PIDs 39422, 39226).
4. Ran fleet audit script confirming 0 residual configs, 0 residual dirs, and 0 running processes (100% clean).
5. Executed live fleet_config_reconcile.zsh to ensure rsync --delete synchronized skill trees across all nodes with current=6 updated=0 diverged=0.
6. Resolved ticket TASK-20260831-34.

Verification Evidence

• Fleet audit results:
  • [rdmsm4x] -> CLEAN (100%)
  • [rdmbair15m5] -> CLEAN (100%)
  • [rdmbair13m5] -> CLEAN (100%)
  • [jdmbair13m5] -> CLEAN (100%)
  • [rdmpw3265m] -> CLEAN (100%)
  • [rdmpw3275m] -> CLEAN (100%)
• Fleet config reconcile report: current=6 updated=0 diverged=0 offline=0 err=0.

Backup Locations

• Pre-mutation backups archived host-locally at ~/Library/Application Support/MCPFleetDisable/backups/stitch-removal-*/ on each machine.

Outstanding Owner Actions

• None.