Allow disabling cloud MCP integrations per-project

Resolved 💬 3 comments Opened Apr 12, 2026 by latticedefect Closed Apr 16, 2026

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

Problem

Cloud MCP integrations (Gmail, Google Calendar, Google Drive, Slack, Linear) are enabled at the account level with no
way to disable them per-project. Every conversation in every project pays the context window cost of injecting all their
tool schemas, even when they're completely irrelevant.

There is no project-level setting, CLI flag, or .mcp.json override that works for cloud integrations — only for locally
configured MCP servers.

Use case

I work on multiple projects. Some are collaborative and benefit from Slack/Linear/Calendar integration. Others (e.g. a
research/thesis project) never use any of these tools. In those projects, the injected schemas are pure context waste
every turn.

The only current workaround is manually disconnecting integrations from the claude.ai account and reconnecting them when
switching projects. That's not a viable workflow.

Proposed solution

A project-level config to disable specific cloud MCP servers, e.g. in .claude/settings.local.json:

{
"disabledMcpServers": ["Gmail", "Google Calendar", "Google Drive", "Slack", "Linear"]
}

Or equivalently, a per-project allowlist so only opted-in servers load.

Why this matters

  • Context window is a finite resource — wasted tokens on unused tool schemas reduce capacity for actual work
  • Local MCP servers already support project-level scoping via .mcp.json — cloud integrations should have parity
  • Users shouldn't have to choose between "available everywhere" and "available nowhere"

Proposed Solution

Problem

Cloud MCP integrations (Gmail, Google Calendar, Google Drive, Slack, Linear) are enabled at the account level with no
way to disable them per-project. Every conversation in every project pays the context window cost of injecting all their
tool schemas, even when they're completely irrelevant.

There is no project-level setting, CLI flag, or .mcp.json override that works for cloud integrations — only for locally
configured MCP servers.

Use case

I work on multiple projects. Some are collaborative and benefit from Slack/Linear/Calendar integration. Others (e.g. a
research/thesis project) never use any of these tools. In those projects, the injected schemas are pure context waste
every turn.

The only current workaround is manually disconnecting integrations from the claude.ai account and reconnecting them when
switching projects. That's not a viable workflow.

Proposed solution

A project-level config to disable specific cloud MCP servers, e.g. in .claude/settings.local.json:

{
"disabledMcpServers": ["Gmail", "Google Calendar", "Google Drive", "Slack", "Linear"]
}

Or equivalently, a per-project allowlist so only opted-in servers load.

Why this matters

  • Context window is a finite resource — wasted tokens on unused tool schemas reduce capacity for actual work
  • Local MCP servers already support project-level scoping via .mcp.json — cloud integrations should have parity
  • Users shouldn't have to choose between "available everywhere" and "available nowhere"

Alternative Solutions

_No response_

Priority

Critical - Blocking my work

Feature Category

CLI commands and flags

Use Case Example

_No response_

Additional Context

_No response_

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗