What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
To see which configured MCP servers have the largest tool definitions, Hao Li’s mcp-tax command-line utility audits stdio servers and estimates their schema-token cost. Its numbers are a rough comparison aid—not exact Claude token counts—and its scope excludes remote HTTP transports and runtime tool results.
Why MCP tools can add context overhead
MCP servers can expose tool definitions for an AI client to discover and pass along to a model. The official Model Context Protocol TypeScript SDK documentation describes tool entries as including names, descriptions, and input schemas. Large or detailed schemas can therefore contribute to the material associated with tool use.
As an Amazon Associate I earn from qualifying purchases.
That does not establish that every MCP host loads every tool definition into every request. The actual behavior depends on the host and its integration. mcp-tax is aimed at a narrower, practical question: among the stdio servers it can inspect, which return the largest serialized tool lists?
Free tools Windows power users keep installed
One-click scans. No signup required.
How mcp-tax measures server schemas
In his September 30, 2026 DEV Community article, tool author Hao Li says mcp-tax reads the user configuration at ~/.claude.json and, if present, the project-level ./.mcp.json. It lists configured servers and audits stdio servers by launching their processes over standard input and output, sending an MCP initialize request, then requesting tools/list.
#1 Best Overall
- More for the money with this high quality Product
- Offers premium quality at outstanding saving
- Excellent product
- 100% satisfaction
The utility serializes the returned tool-list schema compactly, counts its characters, and estimates tokens by dividing that count by four and rounding. This makes the result useful for comparing servers within the same audit, but not a precise count of tokens seen by a model. Tokenizers differ, and Claude Code may wrap schemas in ways the estimate does not capture. Treat the displayed number as an order-of-magnitude gauge, not a billing meter.
What the example output says
Li’s article gives an illustrative example totaling 60 tools and 130,643 schema characters. Applying mcp-tax’s characters-divided-by-four estimator yields 32,661 estimated tokens—about 16.3% of an assumed 200,000-token context window. That is Hao Li’s example calculation from 2026, not an independently reproduced result, representative benchmark, or typical MCP overhead statistic.
| Server in Li’s example | Tools | Schema characters | Estimated tokens at characters ÷ 4 |
|---|---|---|---|
| GitHub | 51 | 118,203 | 29,551 |
| PostgreSQL | 9 | 12,440 | 3,110 |
| Total | 60 | 130,643 | 32,661 |
These figures describe the example in Li’s article, rather than a fresh measurement. The GitHub entry alone accounts for most of its estimated schema size; the comparison is a useful way to spot candidates for closer inspection, not proof of the context cost in another setup.
Disabling servers for a Claude Code session
Li describes controls to mark a server off or on, with disabled server names saved in ~/.config/mcp-tax/disabled.json. The utility’s run command is intended to start Claude Code with a filtered MCP configuration that omits disabled servers, while leaving the original ~/.claude.json unchanged.
Rank #3
- Product type: Screw kit
- Made by Super Micro
- Manufacturer part number: MCP-410-00005-0N
- Supermicro MCP-410-00005-0N Screw Bag(100PCS) and Label for 24x Hot swap
- Mfr Part Number: MCP-410-00005-0N
- Run
mcp-taxto list configured servers and review the reported schema estimates. - Use
mcp-tax off <server>to mark a server disabled, replacing<server>with its configured name. - Launch a filtered session with
mcp-tax run -- <Claude Code arguments>, supplying the Claude Code arguments after--. - To re-enable a server in the utility’s saved state, use
mcp-tax on <server>.
The article notes that Li did not have Claude Code installed to verify the claude --mcp-config launch option. This is a verification caveat, not evidence that the option is unavailable. If the flag does not work with the installed CLI, mcp-tax reportedly prints the filtered configuration path so it can be passed manually.
Coverage limits and what the estimate misses
- Transport: The described audit supports stdio servers, not SSE or streamable HTTP servers.
- Operating system: Li says the
select(2)stdout-pipe handling works on Linux and macOS, not Windows. - Runtime output: The measurement covers returned tool schemas, not the size of results produced when tools run. A small catalog may still return large outputs.
- Token accuracy: The divide-by-four calculation does not use the model’s tokenizer or account exactly for host-specific schema wrapping.
Li reports that seven smoke tests passed, including a fake stdio MCP server using newline-delimited JSON-RPC 2.0 framing, with handshake, stdout-noise, and timeout cases. That is the author’s report; it is not an independent test of the utility.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.When mcp-tax is useful
Use it when you want a quick, same-machine comparison of stdio server tool catalogs and a way to try a Claude Code session with selected servers filtered out. If a server’s estimate looks unusually large, inspect its tool definitions and decide whether you need that server for the task at hand.
Do not use its output as a complete context audit: it does not establish what a particular host includes on each request, cover remote transports, count runtime results, or provide tokenizer-exact usage. The available sources also do not establish a typical MCP context-overhead figure across users or deployments.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
