MCP context budget auditor

See how much context your tools burn before the task starts.

Paste an MCP config, exported tool list, or tools/list response. The auditor estimates overhead, flags tool sprawl, and writes a pruning plan you can use before opening Claude Code, Cursor, or another agent client.

Local-only tool

Audit an MCP setup

Your pasted text is processed in this browser. No config or tool schema is uploaded by this static page.

Paste JSON to start. Non-JSON text is still estimated, but detailed tool analysis needs structured data.

Session assumptions

What it checks

Most context waste is visible before the first prompt.

Tool count

Large MCP servers can expose dozens of tools even when a task only needs three or four.

Description bulk

Long tool descriptions and schemas become hidden context overhead in agent sessions.

Overlap

Duplicate verbs like search, fetch, read, write, and list often indicate tools that should be split by task profile.

Filtering plan

The output helps you decide which server or tool group to keep for coding, research, data, or deployment work.

FAQ

MCP context budget questions

What is an MCP context budget?

It is the portion of an AI agent context window consumed by MCP servers, tool descriptions, schemas, project rules, and related setup before the main task begins.

What does this MCP auditor check?

It checks server count, tool count, estimated token overhead, long descriptions, overlapping tool verbs, large servers, and whether a task-sized profile would be safer.

Is my MCP config uploaded?

No. The current static tool runs in your browser. You should still remove secrets, keys, and private production values before pasting any config into a web page.