Skip to content

chore: optimize MCP tool descriptions and reduce token usage

Joe Burnettrequested to merge
jburnett/mcp-lower-context-overhead into main

Description

I noticed that glab duo claude was compacting conversations quite frequently because of context pressure due to the glab MCP tool descriptions.

This MR optimizes the MCP server implementation to reduce token usage when Claude Code uses GitLab CLI commands through MCP tools.

Changes:

  • Truncate long command descriptions to 100 characters at word boundaries
  • Remove verbose flag descriptions and default values from tool schemas
  • Add server-level usage instructions instead of duplicating guidance per tool
  • Simplify parameter documentation to essential information only
  • Update tests to validate new truncation behavior

Impact:

  • Significantly reduces context overhead for MCP tool usage
  • Maintains functionality while improving token efficiency
  • Preserves essential command information for proper usage

Related Issues

Related to MCP context optimization efforts

How has this been tested?

  • All existing unit tests pass with updated expectations
  • Manual testing of MCP server functionality confirms tools work correctly
  • Verified truncation logic handles various edge cases (no spaces, newlines, etc.)
  • Confirmed server-level instructions are properly included in MCP responses

Screenshots (if appropriate):

Baseline context (no MCP tools):

> /context
  ⎿  ⛁ ⛀ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛀ ⛶
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶   Context Usage
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶   claude-sonnet-4-20250514 • 15k/200k tokens (7%)
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶   ⛁ System prompt: 3.2k tokens (1.6%)
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶   ⛁ System tools: 11.5k tokens (5.7%)
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶   ⛁ Messages: 141 tokens (0.1%)
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶   ⛶ Free space: 185.1k (92.6%)
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶

Baseline context at HEAD with glab MCP:

> /context
  ⎿  ⛁ ⛀ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁   Context Usage
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁   claude-sonnet-4-20250514 • 158k/200k tokens (79%)
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁   ⛁ System prompt: 3.2k tokens (1.6%)
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁   ⛁ System tools: 11.5k tokens (5.7%)
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁   ⛁ MCP tools: 143.5k tokens (71.8%)
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁   ⛁ Messages: 96 tokens (0.0%)
     ⛀ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶   ⛶ Free space: 41.7k (20.8%)
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶

Baseline context after this MR:

> /context
  ⎿  ⛁ ⛀ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁   Context Usage
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁   claude-sonnet-4-20250514 • 97k/200k tokens (48%)
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛀   ⛁ System prompt: 3.3k tokens (1.6%)
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶   ⛁ System tools: 11.5k tokens (5.7%)
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶   ⛁ MCP tools: 81.8k tokens (40.9%)
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶   ⛁ Messages: 8 tokens (0.0%)
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶   ⛶ Free space: 103.4k (51.7%)
     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶

Types of changes

  • Bug fix (non-breaking change which fixes an issue)
  • New feature (non-breaking change which adds functionality)
  • Breaking change (fix or feature that would cause existing functionality to change)
  • Documentation
  • Chore (Related to CI or Packaging to platforms)
  • Test gap

Merge request reports