More screenshots
Works with the AI Agents you already use
MCP Token Budget — What Your Tool Schemas Cost
Reads a captured tools/list and estimates the context your MCP servers consume, which tools are expensive, and how much is schema rather than useful description.
Free
MCP Token Budget — What Your Tool Schemas Cost
Example session with this skill installed
python3 scripts/token_budget.py tools.json --fail-on high
{"servers":[{"name":"notion","tools":[{"name":"notion_fetch_page",
"description":"Fetch a page and its blocks…","inputSchema":{"type":"object","properties":{…}}}]}]}
- Read your context and instructions
- Compiled the mcp token budget
mcp-token-budget 3 servers 43 tools
estimated 53,673 tokens (JSON chars / 4)
SERVER TOOLS TOKENS SCHEMA
notion 25 50,360 97.7%
github 12 2,473 71.9%
ops 6 840 70.9%
HIGH M1 total context cost About 53,673 tokens across 43 tools
HIGH M2 oversized tool notion_fetch_page About 9,162 tokens
MEDIUM M7 schema-heavy server notion 97.7% of its context is input schema
FAIL findings at or above 'high'
Connects securely to your tools. The creator never sees your data.
About this skill
Every MCP server spends context before the user says anything. The tool names, descriptions and JSON schemas are serialized into the prompt, and a schema-heavy server can cost more than the conversation it is meant to support. This puts a number on that cost and shows where it goes.
Point it at a captured tools/list response. It estimates the tokens per server and per tool, ranks the most expensive definitions, and separates schema from prose so you can see whether the cost is structural or explanatory. It flags definitions above 1,000 and 2,000 estimated tokens, servers with more than 20 tools, duplicate names across servers, missing descriptions and very long prose descriptions.
- Per-server and per-tool token estimates from the serialized JSON
- Schema share, because a 97%-schema server is a different fix from a verbose one
- A ranked list of the tools worth splitting or trimming first
- Duplicate tool names that can route a call to the wrong server
- Exit 1 at a severity threshold, so a growing tool registry fails CI
The shipped capture is three servers and 43 tools at roughly 53,700 estimated tokens; one Notion server accounts for about 50,400 of them and a single page-fetch tool is about 9,200. The clean capture is eight tools at about 961 tokens.
Good to know: the estimate is serialized JSON characters divided by four. It is a floor and a relative ranking, not a model tokenizer count; clients add framing, may omit fields and use different tokenizers. A large definition is not automatically bad — the purpose, when-to-use rule and parameter formats are what make selection work. Prune repeated examples and redundant nesting before shortening the boundary text. Cost is not selection quality; pair this with tool-description-lint.
Estimates are based on 4 characters per token. Does not account for specific model tokenizers or client-side framing. Pair with description linting for quality audits.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 day ago
- Free with an Agensi account