[FR] Options to reduce MCP token usage

[kkirss]

Problem

I'm using Task Genius MCP with Claude Code, and the token overhead is eating my context budget.

Tool definition take ~13.6k tokens. The responses also include spaces and newlines between fields.

Solution

Tool filtering – Make it possible to choose which tools are enabled.

Could be as simple as an enabledTools array in settings, filtering what getTools() returns.

Compact JSON – The server uses JSON.stringify(result, null, 2). An option for minified output would be great.

Field selection – Most of my queries don't need all of the fields. Something like query_tasks({ fields: [...] }) or a per-tool config would be useful.

[Quorafind]

Soon. I am also using claude code with task genius

[Quorafind]

Fixed, you can try again with 9.12.3 and see if issue remains

[kkirss]

You added fields parameter to allow filtering fields. This is great, thanks :)

Stripping the whitespace would also be good. The JSON is descriptive enough for LLMs.

It seems you added 2 tools for dynamic discovery of tools. But all the other tools are still included in the context window. So the context usage actually increased with this.

[yifanyifan897645]

The irony of adding discovery tools increasing total context usage is a common trap. If you want to measure the actual token impact of each tool definition before and after changes, mcp-checkup can give exact numbers per tool and grade overall efficiency:

npx mcp-checkup

It would flag a net-negative "optimization" immediately. For compact output, stripping null, 2 formatting from JSON.stringify in tool responses is also low-hanging fruit (~15-20% token reduction).