MCP VSCode diagnostics showing stale/incorrect errors + Request for configurable tool usage

[moncapitaine]

Issue: MCP VSCode Diagnostics Showing Stale/Incorrect Errors

Problem Description

The mcp__ide__getDiagnostics tool is showing TypeScript errors that don't actually exist when verified with the actual TypeScript compiler.

Example Case

While working on a TypeScript project, the MCP diagnostics reported:

✘ [Line 135:62] Property 'addListField' does not exist on type '{}'. [2339] (ts)
✘ [Line 149:65] Property 'updateListField' does not exist on type '{}'. [2339] (ts)

However, running pnpm typecheck showed no TypeScript compilation errors whatsoever. The code was actually correct, but the MCP diagnostics were showing stale or incorrect information.

Impact

• Claude Code makes unnecessary code changes based on incorrect diagnostic information
• Wastes time debugging non-existent issues
• Reduces confidence in the diagnostic system
• Can lead to introducing actual bugs while "fixing" phantom errors

Current Workaround

Cross-verify MCP diagnostic results with actual compilation commands like:

• pnpm typecheck
• npm run lint
• tsc --noEmit

Feature Request: Configurable Tool Usage

Motivation

Different projects and workflows may benefit from different tool configurations. Some tools may be unreliable in certain environments or conflict with project-specific tooling.

Proposed Solution

Add configuration options to allow users/projects to:

1. Disable specific tools: Allow blacklisting problematic tools

   {
     "claude-code": {
       "disabled-tools": ["mcp__ide__getDiagnostics", "mcp__ide__executeCode"]
     }
   }
   

2. Tool reliability preferences: Prefer actual CLI commands over MCP integrations

   {
     "claude-code": {
       "prefer-cli-over-mcp": true,
       "tool-preferences": {
         "typescript-check": "pnpm typecheck",
         "linting": "pnpm lint"
       }
     }
   }
   

3. Tool usage policies: Configure when certain tools should/shouldn't be used

   {
     "claude-code": {
       "tool-policies": {
         "mcp__ide__getDiagnostics": {
           "enabled": false,
           "reason": "Shows stale diagnostics, prefer 'pnpm typecheck'"
         }
       }
     }
   }
   

Configuration Locations

Could be configured via:

• .claude-code.json in project root
• Global settings file
• Environment variables
• CLAUDE.md instructions (already partially supported)

Benefits

• Project-specific workflows: Teams can optimize Claude Code for their specific toolchain
• Reliability: Disable unreliable tools in favor of proven alternatives
• Performance: Avoid expensive or slow tools when faster alternatives exist
• Debugging: Easier to troubleshoot by systematically enabling/disabling tools

Examples of Use Cases

• Disable MCP diagnostics in favor of CLI TypeScript checking
• Prefer project-specific test runners over generic execution
• Skip certain linting tools that conflict with project setup
• Use custom build/deployment scripts instead of generic ones

This would make Claude Code much more adaptable to different development environments and increase reliability by allowing users to work around problematic integrations.

Environment

• Working directory: Node.js project with TypeScript
• MCP tools: VSCode integration enabled
• TypeScript: Working correctly via CLI, but MCP diagnostics showing false errors

Would appreciate both fixing the diagnostic staleness issue and considering the configurable tool usage feature for better workflow customization.

---

🤖 This issue was generated with Claude Code

[marcospgp]

I found a similar issue where issues in files ignored by VS Code which are not reported in the problems pane are reported by this MCP tool call.

https://github.com/anthropics/claude-code/issues/8963#issuecomment-3368995026

[github-actions[bot]]

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

[marcospgp]

keep alive