Skip to content

[deep-report] MCP get_file_contents has no excerpt/range-read mode — full-file reads waste ~4,500 tokens on large files #50594

Description

@github-actions

Description

The MCP Structural Analysis has flagged the GitHub MCP get_file_contents tool as the lowest-rated tool tested (2/5) for two consecutive days (2026-08-04 and 2026-08-05): it only supports whole-file reads with no excerpt/range mode. A single README fetch cost ~4,200-4,800 tokens, dominated by boilerplate (e.g. a ~300-entry contributor list) before reaching useful content.

Expected Impact

Meaningfully reduces token bloat for any agent that needs to read a small section of a large file — a recurring, cross-workflow cost given how frequently file-content reads happen across the fleet.

Suggested Agent

New Agent / MCP tooling maintainer — add a line-range or byte-offset parameter to the tool's request schema, or a wrapper (similar to the mcpscripts pattern already used for list_workflows) that supports partial reads.

Estimated Effort

Medium (1-4 hours).

Data Source

DeepReport Intelligence Briefing analysis, 2026-08-05, sourced from MCP Structural Analysis — 2026-08-05 (carried over from the 2026-08-04 edition of the same report).

Generated by 🔬 Deep Report · agent · 160.4 AIC · ⌖ 36.9 AIC · ⊞ 11K ·

  • expires on Aug 7, 2026, 7:34 AM UTC-08:00

Metadata

Metadata

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions