Docs for AI coding agents
Singapore AI showcase
22 Apr 2025
Vivek Kalyan
hello@vivekkalyan.com
About Me
AI Research Lead @ Handshakes (2018 - 2024)
- Knowledge Graphs (Corporate Entities, Events), Document Classification, Information Extraction, Recommendation Systems, Local LLMs
Founder @ Cartograph (2024 - Present)
- Knowledge Graphs (Code), Code Understanding, LLMs, Tools, Agents
Cartograph
Code documentation platform for software teams
Problem
AI models code understanding/generation
- works best with popular libraries, frameworks, languages
- not as good for less popular/new technology
- impossible to work for internal libraries
Stifles innovation, new ideas
Most of the world's code is written in boring technology
Provide LLMs hosted, clean markdown documentation of libraries and frameworks
- Processes and hosts documentations from various libraries
- Converts HTML pages to clean markdown
- Removes unnecessary content (headers, sidebar, footers)
- Standardizes links - easy for agents to get the right page
- Backoff logic to handle errors
Model Context Protocol Server:
list_docs: List all available documentation setsget_docs_index: Retrieves a condensed, LLM-friendly index of a documentation setget_docs_full: Retrieves a complete documentation set in a single consolidated filesearch_docs: Search a documentation set by keywordsget_docs_page: Retrieves a specific page of documentation set
DEMO
{
"mcpServers": {
"atlas-docs": {
"command": "npx",
"args": ["-y", "@cartographai/atlas-docs-mcp"]
}
}
}Why can't agents find the information themselves?
Documentation
- Browser use (slow, expensive) - for now at least
- Redundant Effort (time/token cost to scrape, index, search sites)
- Fragile - 404s, misformated links, server errors
- HTML is not the right abstraction for documentation
- HTML: 37k tokens, Markdown: 1k tokens
Code
- Exponential sub-dependencies (especially in JS land)
Try it out!
Try it with any MCP compliant client: Cursor, Windsurf, Cline, Claude Code
Let us know what documentation you want to be hosted
Thank you!
Tools for AI coding agents
By vivekkalyan
