Skip to content

get_board_info response is too large, consuming excessive LLM context tokens #218

@sadjow

Description

@sadjow

Problem

The get_board_info tool returns an extremely large response (~15k+ tokens) for boards with many columns and groups. When used inside an LLM agent (e.g., Claude Code), this fills up the context window quickly, leaving less room for actual task work.

The response includes deeply nested column settings, full label definitions with colors/hex codes, and all group metadata — most of which is not needed for the majority of use cases.

Example

A board with ~25 columns and ~11 groups produces a ~68KB JSON response. The MCP client warns:

⚠ Large MCP response (~15.3k tokens), this can fill up context quickly

Suggestion

Consider one or more of the following:

  1. Add a fields or include parameter to let callers request only specific parts of the board info (e.g., only columns, only groups, only labels for specific columns).
  2. Reduce default verbosity — e.g., omit column settings details by default, or omit color/hex metadata from status labels unless explicitly requested.
  3. Add pagination or summary mode — return a compact summary by default with an option to drill into specific columns or groups.
  4. Truncate or cap response size — set a reasonable token budget and summarize when exceeded.

This would make the tool much more practical for LLM-based agents where context window management is critical.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions