PulseAugur
EN
LIVE 23:04:22

MCP servers' token costs vary wildly, with some exceeding 200k context windows

A recent analysis of 35,633 MCP servers revealed that while the median server is relatively inexpensive in terms of token cost, a significant long tail of servers consumes a disproportionately large amount of context window space. The median server costs 2,518 tokens, but the average is over three times that due to a few extremely large configurations. Approximately 45 servers have tool definitions alone that exceed a 200,000 token context window, with the largest consuming 1.6 million tokens. The primary driver for this high token cost is the sheer number of tools a server defines, rather than the complexity of individual tool schemas. AI

IMPACT Highlights potential inefficiencies in how large language model tools are defined and managed, impacting context window usage and cost.

RANK_REASON Analysis of token costs for a large number of servers, presenting new data and findings. [lever_c_demoted from research: ic=1 ai=1.0]

Read on dev.to — MCP tag →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

MCP servers' token costs vary wildly, with some exceeding 200k context windows

How we ranked this

Signal score
41 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
Analysis of token costs for a large number of servers, presenting new data and findings. [lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
infra
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

Full methodology in our editorial standards.

COVERAGE [1]

  1. dev.to — MCP tag TIER_1 English(EN) · ondrej ·

    I counted the exact token cost of 35,633 MCP servers. 45 of them do not fit in a 200k context window

    <p>An MCP server costs you tokens before you ask it anything. The client sends the model every tool definition, name, description and schema, and those definitions take space in every request of the session, used or not.</p> <p>Last month Sarvar Nadaf wrote a good post about this…