Reduce token usage in MCP tools by avoiding unnecessary data fetching

[RomKadria]

Summary

Three targeted optimizations to reduce response payload sizes in platform API tools, cutting unnecessary token consumption:

1. get_updates — Conditional fetching + dropped HTML body

Problem: The GraphQL queries always fetched replies and assets from the monday.com API, even when includeReplies and includeAssets were false. The tool discarded them in JavaScript after they'd already been transferred. Additionally, every update returned both body (HTML markup) and text_body (plain text) — the HTML is unnecessary for LLM consumption and roughly doubles each update's content size.

Fix:

• Added @include(if: $includeReplies) and @include(if: $includeAssets) directives to both GetItemUpdates and GetBoardUpdates GraphQL queries, so the API itself skips these fields when not requested.
• Removed the body (HTML) field from queries and response mapping — text_body is sufficient for LLM tools.

2. get_board_activity — New includeData option (default false)

Problem: Each activity log entry has a data field containing the full raw JSON of before/after column state for every change. With the default limit of 1000 entries, this produces massive payloads — often 100k+ tokens of raw JSON that the LLM rarely needs.

Fix:

• Added an includeData input parameter (boolean, default false) with a descriptive schema.
• Added @include(if: $includeData) directive to the data field in the GraphQL query so it's not even fetched when not needed.

3. list_users_and_teams — Fixed listUsersOnly query

Problem: The listUsersOnly GraphQL query was identical to listUsersWithTeams — both fetched teams { ...UserTeamMembership } for every user. With DEFAULT_USER_LIMIT = 1000, this meant team membership data for up to 1000 users was being fetched and serialized for no reason on the "users only" code path.

Fix: Removed the teams fragment from listUsersOnly so it actually matches its semantic intent.

Test plan

• All 13 get-updates-tool tests pass (including updated assertions for new includeReplies/includeAssets variables)
• All 14 list-users-and-teams-tool tests pass
• GraphQL codegen succeeds with updated queries
• Zero linter errors on all changed files

Made with Cursor

[damian-rakus]

Before merging let's make sure all relevant subgraphs support the @include

[RomKadria]

Before merging let's make sure all relevant subgraphs support the @include

Yeah I sent them a message, I think you are in that channel as well