AI Knowledge Format Comparison
OKF vs MCP vs AGENTS.md vs llms.txt

Overview: The AI Knowledge Landscape in 2026

The way AI agents consume knowledge is rapidly standardizing. In 2026, four major formats have emerged as the leading approaches for making information accessible to AI systems — each solving a different part of the puzzle. Whether you're building developer tools, running an open-source project, or managing enterprise documentation, understanding these formats is essential for making your content truly AI-ready.

Here's the key insight: these formats are not competitors — they're complementary layers in the AI knowledge stack. OKF provides deep, structured domain knowledge. MCP enables real-time tool interaction. AGENTS.md gives coding agents project-level instructions. llms.txt helps AI systems discover what's available on your site. Used together, they create a complete AI accessibility strategy.

This guide compares all four formats across key dimensions — what they are, how they work, their strengths and weaknesses, and most importantly, which format (or combination) is right for your specific use case.

OKF — Open Knowledge Format

What It Is

OKF (Open Knowledge Format) is a directory-based standard for packaging structured domain knowledge as portable markdown bundles. Launched by Google Cloud Platform in mid-2026, OKF defines how knowledge should be organized, annotated, and version-controlled for direct consumption by AI agents. Each bundle is a self-contained directory of markdown files with YAML frontmatter metadata, cross-references, and a standardized directory structure.

How It Works

An OKF bundle is a directory tree — typically version-controlled in a git repository — containing an index.md manifest, categorized markdown files in subdirectories (concepts, guides, reference, tutorials), and a log.md changelog. Every file includes YAML frontmatter with title, description, tags, version, and status metadata. AI agents load the entire bundle into context, gaining deep, structured understanding of the domain without needing to browse websites or fetch additional pages.

OKF bundles are indexed on BundleDex, the definitive directory with hundreds of curated bundles, making them discoverable to agents across the ecosystem.

What It's Best For

• Deep domain knowledge — API references, architecture guides, SDK documentation
• Portable knowledge — bundles work across any agent platform, IDE, or tool
• Offline-first environments — zero infrastructure needed, plain markdown files
• Version-controlled knowledge — git-native with full change history, PRs, and reviews
• Cross-referenced documentation — concepts link to each other, forming a knowledge graph

Learn more: What Is OKF? | OKF vs MCP | OKF vs llms.txt

MCP — Model Context Protocol

What It Is

MCP (Model Context Protocol) is an open protocol developed by Anthropic that standardizes how AI agents interact with external tools and services in real time. Unlike OKF (which is about static knowledge), MCP is about live action — querying databases, calling APIs, searching documentation, and executing operations during an agent session. It uses a client-server architecture with JSON-RPC transport over stdio or SSE.

How It Works

An MCP server exposes a set of tools, each defined with a JSON Schema that describes its inputs and outputs. When an AI agent (MCP client) connects to a server, it performs a handshake to discover available tools and their schemas. During a session, the agent dynamically selects and calls tools as needed — for example, searching a database, checking deployment status, or querying a live API. Results are returned in real time, enabling agents to make decisions based on current data.

What It's Best For

• Live data access — database queries, API calls, system status checks
• Action execution — creating tickets, deploying code, sending notifications
• Real-time search — semantic search over large, dynamic document corpora
• Tool orchestration — chaining multiple operations in a single agent workflow
• Dynamic computation — calculations, data transformations, code execution

Learn more: OKF vs MCP | BundleDex MCP Server

AGENTS.md — Project-Level Agent Instructions

What It Is

AGENTS.md is a convention for providing project-level instructions to AI coding agents. A single markdown file placed in the root of a repository, it tells AI coding assistants (like Cursor, Claude Code, GitHub Copilot, and others) how to work with the project — coding conventions, build commands, testing procedures, architecture overview, and project-specific guidelines. It's the simplest of the four formats: one file, no tooling required.

How It Works

When an AI coding agent opens a project, it automatically looks for an AGENTS.md (or .cursorrules) file in the repository root. The file contains markdown instructions that the agent reads into its context — explaining the project structure, coding standards, testing workflow, and any special considerations. The agent uses these instructions to make better decisions about code generation, refactoring, and debugging within that specific codebase.

What It's Best For

• Coding agent guidance — project conventions, build commands, test procedures
• Quick setup — one file, zero dependencies, works instantly with most coding agents
• Team consistency — ensures all developers' AI assistants follow the same rules
• Onboarding — gives new contributors' AI agents immediate project context
• Complementing OKF — AGENTS.md provides agent instructions; OKF provides detailed domain knowledge

Learn more: AGENTS.md vs Cursor Rules vs OKF

llms.txt — Website-Level AI Discovery

What It Is

llms.txt is a lightweight discovery standard that helps AI systems find and understand what content is available on a website. Proposed by Jeremy Howard in late 2024, it's a single markdown file placed at the well-known URL /llms.txt that lists links to the site's AI-readable resources — documentation, API references, guides, and knowledge bundles. It's the AI equivalent of robots.txt, but for content discovery rather than crawling rules.

How It Works

When an AI agent or LLM-based tool encounters a website, it can check /llms.txt to discover what structured content is available. The file contains a brief description of the site followed by a list of links to markdown, JSON, or other machine-readable resources. Some sites also provide a /llms-full.txt with the complete content in a single file. The format is deliberately simple — no schema, no tooling, just a markdown file with links.

What It's Best For

• Website discovery — tells AI agents what documentation and resources are available
• SEO for AI — makes your content discoverable to LLMs and AI search tools
• Minimal effort — one markdown file, easy to create and maintain
• Complementing OKF — llms.txt points agents to your OKF bundles and other resources
• Broad compatibility — supported by a growing number of AI platforms and tools

Learn more: OKF vs llms.txt

Comparison Matrix

Here's how all four formats compare across the dimensions that matter most for AI agent knowledge:

Decision Guide: Which Format for Which Use Case

Use OKF When...

• You have comprehensive documentation that agents need to understand deeply
• You want portable, offline-capable knowledge that works across any platform
• You need version-controlled, auditable knowledge artifacts
• You're publishing SDK documentation, API references, or architecture guides
• You want your knowledge discoverable on BundleDex

Use MCP When...

• You have live data or services agents need to query in real time
• You want agents to perform actions — deploy, query, create, update
• Your data changes frequently and can't be pre-packaged as static files
• You're building database-backed tools or live API integrations

Use AGENTS.md When...

• You want AI coding agents to follow project-specific conventions
• You need a quick, zero-dependency way to guide coding assistants
• Your team uses Cursor, Claude Code, Copilot, or similar AI coding tools
• You want to complement your OKF bundles with agent-specific instructions

Use llms.txt When...

• You have a documentation website and want AI agents to discover it
• You want a lightweight entry point that links to your OKF bundles and docs
• You need the simplest possible AI discovery mechanism
• You want to be part of the growing llms.txt ecosystem

FAQ