AI agents now account for a growing share of developer docs traffic

AI agents are now responsible for a substantial share of reads on developer documentation hosted by Mintlify, with Claude Code, ChatGPT and Claude accounting for 88% of agent reads over the last 90 days. Mintlify says weekly searches through docs MCP servers also tripled during the period.
- Claude Code generated 40% of agent documentation reads.
- Claude Code, ChatGPT and Claude accounted for 88% of agent reads.
- Weekly docs MCP searches rose from 262K to 812K.
- AI product documentation received 55% of reads across the top 300 sites.
- Mintlify says the data undercounts agents using generic tools.
Three agents dominate documentation reads
Mintlify classified hundreds of millions of documentation reads from August to November 2026 using agent-traffic, its open-source dataset of agents and crawlers. The analysis covers documentation sites hosted by Mintlify, which says it hosts docs for thousands of companies.
Claude Code generated 40% of agent documentation reads and read 92% of the documentation sites included in the reach analysis. ChatGPT and Codex, which OpenAI sends under the same user agent, accounted for 37% of reads and reached 86% of sites. Claude, referring to the Claude app and API rather than Claude Code, made up 10% of reads and reached 93% of sites.
All three agents read 84% of the active documentation sites measured by Mintlify. OpenCode reached 59% of sites but generated less than 1% of reads, a difference Mintlify says suggests broad but relatively light use.
Coding agents and docs searches are expanding
Mintlify says several coding agents increased their weekly documentation reads during the 90 day period. DeepSeek's harness rose from almost nothing in August to 184K reads a week. OpenCode increased from 3K to 249K, Kiro grew from 569 to 7K, and GitHub Copilot grew from 65K to 254K. The company cautions that an increase can partly reflect an agent beginning to identify itself, rather than beginning to read documentation.
Weekly searches through Mintlify docs MCP servers rose from 262K to 812K, while the number of sites searched each week nearly doubled to about 2,000. Agent queries also differed from human searches. The median agent search was seven words and often read like a specification, while the median human search was one word. Nearly a quarter of agent searches were longer than ten words.
AI products attract most agent documentation reads
Mintlify sorted the 300 documentation sites most read by agents into eight categories. Documentation for AI products received 55% of agent reads on those sites, according to the company, indicating that many projects built with agents rely on other AI products.
Creative AI tools showed the fastest growth, at 3.4x, followed by payments at 3.3x. AI models and inference grew 2.6x, developer infrastructure and business and communication APIs each grew 2.5x, crypto and prediction markets grew 2.4x, and AI coding and app builders grew 2.1x. Agent frameworks and tooling, the largest category, grew 2.0x.
The figures exclude training crawlers, search indexers and generic HTTP clients. Mintlify also says the totals undercount agents that use generic tools such as curl, including Codex shell commands and most of Cursor's traffic. The analysis uses aggregated data, so no individual documentation site is named.
Mintlify: Who reads developer docs now


