Graphify.md · community-built · atlas MCP

AI developer platforms
as traversable knowledge graphs

Eleven AI developer and agent platforms as typed, source-pinned graphs. Search concepts, compare vocabulary across vendors, inspect MCP touchpoints, traverse declared relationships, and verify source hashes from one local MCP server.

Not a general-purpose semantic search layer - if it's not a declared edge, the graph doesn't return it.

$pip install ckg-ai-platforms

"The graph doesn't guess - it traverses. Every answer traces to a declared edge."

PyPI v0.3.0 11 domains · 475 nodes · 550 typed edges local stdio MCP KRB v0.6.2 MIT Graphify.md

PyPI GitHub Benchmark paper HF dataset graphifymd.com Graphify.md

Eleven platform graphs

475 nodes · 550 typed CSV edges · hover to explore sample vocabulary · colored by platform family

LangChain LangGraph OpenAI Anthropic Hugging Face Bedrock Vertex / Gemini / Microsoft / Palantir

What's in the graph

LangChain

48 nodes · 50 edges

Chains · Agents · LCEL · RunnableSequence · Memory · Retrievers · LangSmith · Output Parsers

LangGraph

42 nodes · 51 edges

Graph State · StateGraph · Nodes · Edges · Cycles · Persistence · Checkpointing · Human-in-the-Loop

OpenAI API

43 nodes · 45 edges

Chat Completions · Assistants · Embeddings · DALL-E · Whisper · Function Calling · Tool Use · Batch API

Anthropic SDK

44 nodes · 43 edges

Messages API · Tool Use · Prompt Caching · Computer Use · Extended Thinking · Streaming · Batch Processing

Hugging Face

46 nodes · 49 edges

Hub · Transformers · Spaces · Datasets · Inference API · PEFT · Accelerate · Safetensors

AWS Bedrock

46 nodes · 48 edges

Converse API · Guardrails · Knowledge Bases · Agents · Model Catalog · Prompt Management · Titan Embeddings

Vertex AI

45 nodes · 48 edges

Model Garden · Gemini API · Evaluation · Fine-tuning · Pipelines · Feature Store · Vertex AI Search

AWS AgentCore

36 nodes · 46 edges

Runtime · Gateway · Memory · Identity · Observability · Strands · MCP · RAG touchpoints

Gemini Agent Platform

40 nodes · 46 edges

Vertex AI · Agent Engine · ADK · MCPToolset · A2A · Memory Bank · Evaluation · Deployment

Microsoft AI Agents

40 nodes · 50 edges

Azure AI Foundry · Semantic Kernel · MCP plugins · Agent Service · Tools · Memory · Governance

Palantir Foundry

45 nodes · 74 edges

Ontology · Object Types · Link Types · Action Types · AIP · Datasets · Provenance · Governance

Traversal examples

list_atlas()
11 domains, 475 nodes, 550 typed CSV edges

find_mcp_touchpoints()
Returns MCP concepts in AWS AgentCore, Google ADK, Microsoft Semantic Kernel,
Anthropic SDK, and related platform graphs.

get_concept_context("Palantir Ontology", "palantir-foundry")
Returns direct outgoing and incoming typed relationships plus source hash.

compare_platforms("memory")
Shows which packaged platforms declare memory-related concepts.

Install

# pip
pip install ckg-ai-platforms

# Claude Desktop
{ "mcpServers": {
  "ai-platforms": {
    "command": "uvx",
    "args": ["ckg-ai-platforms"]
  }
}}

# Python library
from ckg_ai_platforms import compare_platforms, get_concept_context
matches = compare_platforms("MCP")
context = get_concept_context("AgentCore Gateway", "aws-bedrock-agentcore")

Tools

Tool	Description
list_atlas()	All packaged platform graphs with counts and source coverage
compare_platforms(query)	Search one term across all vendor/platform graphs
find_mcp_touchpoints()	Locate MCP-related concepts and direct relationships
get_concept_context(concept, domain)	Source-pinned direct typed relationships
query_ckg(concept, domain, depth)	Typed subgraph around any concept, 1-5 hops
get_prerequisites(concept, domain)	Compatibility tool for upstream relationship chains
search_concepts(query, domain)	Fuzzy search within any domain
ask_platform(question, domain)	Natural language query with domain auto-detection

Benchmark (KRB v0.6.2 locked)

0.471

CKG Macro F1

separate public benchmark

0.123

RAG baseline

2,982 tokens · $76.23/1k

3.8x

Macro F1 vs RAG

v0.6.2 structural harness

Try it now

11 platform graphs. One install. No API key.

Give an agent a small map before it reads pages of vendor prose. The server helps it find platform concepts, compare vocabulary, inspect MCP touchpoints, traverse declared relationships, and verify sources.

View on GitHub View on PyPI Read the paper

Pattern: public docs to source-pinned graph to MCP tools.