ckg-ai-platforms 0.1.0__tar.gz

ckg_ai_platforms-0.1.0/.gitignore

@@ -0,0 +1,6 @@
+__pycache__/
+*.py[cod]
+dist/
+*.egg-info/
+.venv/
+.env

ckg_ai_platforms-0.1.0/PKG-INFO

@@ -0,0 +1,82 @@
+Metadata-Version: 2.4
+Name: ckg-ai-platforms
+Version: 0.1.0
+Summary: AI developer platforms as traversable knowledge graphs — LangChain, LangGraph, OpenAI, Anthropic, HuggingFace, AWS Bedrock, Vertex AI — MCP-native
+Project-URL: Homepage, https://graphifymd.com
+Project-URL: Repository, https://github.com/Yarmoluk/ckg-ai-platforms
+Project-URL: Bug Tracker, https://github.com/Yarmoluk/ckg-ai-platforms/issues
+Author-email: Daniel Yarmoluk <daniel.yarmoluk@gmail.com>
+License: MIT
+Keywords: ai-agents,anthropic,bedrock,huggingface,knowledge-graph,langchain,mcp,openai,vertex-ai
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.11
+Requires-Dist: mcp[cli]>=1.0.0
+Description-Content-Type: text/markdown
+
+# ckg-ai-platforms
+
+AI developer platforms as traversable knowledge graphs — MCP-native.
+
+**7 domains · 314 nodes · 445 edges · 11× fewer tokens than RAG**
+
+| Domain | Nodes | Edges | Coverage |
+|---|---|---|---|
+| langchain | 48 | 71 | Chains, agents, LCEL, memory, retrievers, LangSmith |
+| langgraph | 42 | 58 | Graph state, nodes, edges, cycles, persistence |
+| openai-api | 43 | 59 | Chat completions, assistants, embeddings, DALL-E |
+| anthropic-sdk | 44 | 62 | Messages API, tool use, prompt caching, computer use |
+| huggingface | 46 | 65 | Hub, Transformers, Spaces, Datasets, Inference API |
+| aws-bedrock | 46 | 67 | Converse API, Guardrails, Knowledge Bases, agents |
+| vertex-ai | 45 | 63 | Model Garden, Gemini API, evaluation, fine-tuning |
+
+## Install
+
+```bash
+pip install ckg-ai-platforms
+uvx ckg-ai-platforms   # MCP server
+```
+
+## Use as MCP server
+
+Add to your Claude Desktop or any MCP client:
+
+```json
+{
+  "mcpServers": {
+    "ai-platforms": {
+      "command": "uvx",
+      "args": ["ckg-ai-platforms"]
+    }
+  }
+}
+```
+
+## Use as Python library
+
+```python
+from ckg_ai_platforms import ask_platform, list_domains
+
+# Ask a question — auto-detects platform
+result = ask_platform("How does LCEL RunnableSequence work?")
+print(result)
+
+# Explicit domain
+result = ask_platform("explain tool use", domain="anthropic-sdk")
+```
+
+## Benchmark
+
+Part of the [CKG benchmark](https://github.com/Yarmoluk/ckg-benchmark/blob/main/paper/main.pdf):
+- F1: 0.471 vs RAG 0.123 (3.8×)
+- Tokens/query: 269 vs 2,982 (11×)
+- Cost/correct answer: $7.81 vs $76.23
+
+## License
+
+MIT · [graphifymd.com](https://graphifymd.com)

ckg_ai_platforms-0.1.0/README.md

@@ -0,0 +1,61 @@
+# ckg-ai-platforms
+
+AI developer platforms as traversable knowledge graphs — MCP-native.
+
+**7 domains · 314 nodes · 445 edges · 11× fewer tokens than RAG**
+
+| Domain | Nodes | Edges | Coverage |
+|---|---|---|---|
+| langchain | 48 | 71 | Chains, agents, LCEL, memory, retrievers, LangSmith |
+| langgraph | 42 | 58 | Graph state, nodes, edges, cycles, persistence |
+| openai-api | 43 | 59 | Chat completions, assistants, embeddings, DALL-E |
+| anthropic-sdk | 44 | 62 | Messages API, tool use, prompt caching, computer use |
+| huggingface | 46 | 65 | Hub, Transformers, Spaces, Datasets, Inference API |
+| aws-bedrock | 46 | 67 | Converse API, Guardrails, Knowledge Bases, agents |
+| vertex-ai | 45 | 63 | Model Garden, Gemini API, evaluation, fine-tuning |
+
+## Install
+
+```bash
+pip install ckg-ai-platforms
+uvx ckg-ai-platforms   # MCP server
+```
+
+## Use as MCP server
+
+Add to your Claude Desktop or any MCP client:
+
+```json
+{
+  "mcpServers": {
+    "ai-platforms": {
+      "command": "uvx",
+      "args": ["ckg-ai-platforms"]
+    }
+  }
+}
+```
+
+## Use as Python library
+
+```python
+from ckg_ai_platforms import ask_platform, list_domains
+
+# Ask a question — auto-detects platform
+result = ask_platform("How does LCEL RunnableSequence work?")
+print(result)
+
+# Explicit domain
+result = ask_platform("explain tool use", domain="anthropic-sdk")
+```
+
+## Benchmark
+
+Part of the [CKG benchmark](https://github.com/Yarmoluk/ckg-benchmark/blob/main/paper/main.pdf):
+- F1: 0.471 vs RAG 0.123 (3.8×)
+- Tokens/query: 269 vs 2,982 (11×)
+- Cost/correct answer: $7.81 vs $76.23
+
+## License
+
+MIT · [graphifymd.com](https://graphifymd.com)

ckg_ai_platforms-0.1.0/pyproject.toml

@@ -0,0 +1,34 @@
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[project]
+name = "ckg-ai-platforms"
+version = "0.1.0"
+description = "AI developer platforms as traversable knowledge graphs — LangChain, LangGraph, OpenAI, Anthropic, HuggingFace, AWS Bedrock, Vertex AI — MCP-native"
+readme = "README.md"
+license = { text = "MIT" }
+authors = [{ name = "Daniel Yarmoluk", email = "daniel.yarmoluk@gmail.com" }]
+keywords = ["langchain", "openai", "anthropic", "knowledge-graph", "mcp", "ai-agents", "huggingface", "bedrock", "vertex-ai"]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+]
+requires-python = ">=3.11"
+dependencies = ["mcp[cli]>=1.0.0"]
+
+[project.scripts]
+ckg-ai-platforms = "ckg_ai_platforms.server:main"
+
+[project.urls]
+Homepage = "https://graphifymd.com"
+Repository = "https://github.com/Yarmoluk/ckg-ai-platforms"
+"Bug Tracker" = "https://github.com/Yarmoluk/ckg-ai-platforms/issues"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/ckg_ai_platforms"]

ckg_ai_platforms-0.1.0/src/ckg_ai_platforms/__init__.py

@@ -0,0 +1,16 @@
+"""ckg-ai-platforms — AI developer platforms as traversable knowledge graphs."""
+__version__ = "0.1.0"
+
+from ckg_ai_platforms.server import (  # noqa: F401
+    ask_platform,
+    list_domains,
+    search_concepts,
+    query_ckg,
+    get_prerequisites,
+)
+from ckg_ai_platforms.graph import (  # noqa: F401
+    available_domains,
+    load_graph,
+    find_concept,
+    prerequisite_chain,
+)

ckg_ai_platforms-0.1.0/src/ckg_ai_platforms/__main__.py

@@ -0,0 +1,4 @@
+from ckg_ai_platforms.server import main
+
+if __name__ == "__main__":
+    main()

ckg_ai_platforms-0.1.0/src/ckg_ai_platforms/domains/anthropic-sdk.csv

@@ -0,0 +1,45 @@
+ConceptID,ConceptLabel,Dependencies,TaxonomyID
+1,Anthropic SDK,,Platform
+2,Anthropic Client,1,Core
+3,AsyncAnthropic Client,1,Core
+4,Messages API,2,Core
+5,system message,4,Message
+6,user message,4,Message
+7,assistant message,4,Message
+8,content block,4,Message
+9,tool_use block,8,Tool
+10,tool_result block,9,Tool
+11,tool definition,4,Tool
+12,input_schema,11,Tool
+13,computer_use tool,9,Tool
+14,bash_tool,9,Tool
+15,text_editor_tool,9,Tool
+16,tool_choice,11,Tool
+17,Extended Thinking,4,Reasoning
+18,thinking block,17,Reasoning
+19,budget_tokens,17,Reasoning
+20,redacted_thinking,17,Reasoning
+21,Prompt Caching,4,Efficiency
+22,cache_control ephemeral,21,Efficiency
+23,cache_control persistent,21,Efficiency
+24,cache_read_input_tokens,21,Efficiency
+25,Streaming,4,Streaming
+26,stream context manager,25,Streaming
+27,on_message handler,25,Streaming
+28,on_content_block_delta,25,Streaming
+29,Message Batches API,1,Batching
+30,batch request,29,Batching
+31,batch result,29,Batching
+32,Files API,1,Files
+33,file upload,32,Files
+34,file reference,33,Files
+35,Token Counting,4,Efficiency
+36,count_tokens,35,Efficiency
+37,claude-opus-4,1,Model
+38,claude-sonnet-5,1,Model
+39,claude-haiku-4-5,1,Model
+40,MCPServerStdio,1,MCP
+41,MCPServerSSE,40,MCP
+42,RateLimitError,2,ErrorHandling
+43,APIStatusError,42,ErrorHandling
+44,retry with backoff,42,ErrorHandling

ckg_ai_platforms-0.1.0/src/ckg_ai_platforms/domains/aws-bedrock.csv

@@ -0,0 +1,47 @@
+ConceptID,ConceptLabel,Dependencies,TaxonomyID
+1,AWS Bedrock,,Platform
+2,Model Access,1,Foundation
+3,Anthropic Claude,2,Foundation
+4,Amazon Titan,2,Foundation
+5,Meta Llama,2,Foundation
+6,Mistral AI,2,Foundation
+7,Cohere Command,2,Foundation
+8,Stability AI,2,Foundation
+9,InvokeModel API,1,Inference
+10,Converse API,9,Inference
+11,InvokeModelWithResponseStream,9,Inference
+12,ConverseStream,10|11,Inference
+13,Titan Embeddings,4,Embedding
+14,Cohere Embed,7,Embedding
+15,Knowledge Base,1,RAG
+16,Vector Store,15,RAG
+17,OpenSearch Serverless,16,RAG
+18,Aurora PostgreSQL,16,RAG
+19,Pinecone,16,RAG
+20,RetrieveAndGenerate API,15,RAG
+21,Bedrock Agent,1,Agent
+22,Action Group,21,Agent
+23,Lambda Function,22,Agent
+24,Agent Alias,21,Agent
+25,Orchestration Strategy,21,Agent
+26,Return of Control,21|22,Agent
+27,Guardrail,1,Safety
+28,Content Filter,27,Safety
+29,Topic Filter,27,Safety
+30,PII Redaction,27,Safety
+31,Grounding Check,27,Safety
+32,Word Filter,27,Safety
+33,Fine-tuning Job,1,Training
+34,Training Data S3,33,Training
+35,Continued Pre-training,33,Training
+36,Model Customization,33|35,Training
+37,Model Evaluation,1,Evaluation
+38,LLM-as-Judge,37,Evaluation
+39,Human Evaluation,37,Evaluation
+40,Prompt Management,1,Prompt
+41,Prompt Version,40,Prompt
+42,Prompt Flow,40,Prompt
+43,VPC Endpoint,1,Security
+44,KMS Encryption,1,Security
+45,IAM Role,1,Security
+46,CloudTrail Logging,1,Security

ckg_ai_platforms-0.1.0/src/ckg_ai_platforms/domains/huggingface.csv

@@ -0,0 +1,47 @@
+ConceptID,ConceptLabel,Dependencies,TaxonomyID
+1,HuggingFace Hub,,Platform
+2,Model Hub,1,Hub
+3,Dataset Hub,1,Hub
+4,Space Hub,1,Hub
+5,Model Card,2,Hub
+6,LFS Large File Storage,2,Hub
+7,AutoModel,2,Transformers
+8,AutoTokenizer,2,Transformers
+9,pipeline API,7|8,Transformers
+10,Trainer,7,Transformers
+11,TrainingArguments,10,Transformers
+12,BPE Tokenizer,8,Tokenizer
+13,WordPiece Tokenizer,8,Tokenizer
+14,SentencePiece,8,Tokenizer
+15,Fast Tokenizer,8|12,Tokenizer
+16,LoRA,7,PEFT
+17,QLoRA,16,PEFT
+18,PromptTuning,7,PEFT
+19,IA3,7,PEFT
+20,AdaLoRA,16,PEFT
+21,PeftModel,16|17,PEFT
+22,SFTTrainer,10,TRL
+23,DPOTrainer,10,TRL
+24,PPOTrainer,10,TRL
+25,RewardTrainer,10,TRL
+26,InferenceClient,1,Inference
+27,InferenceEndpoint,26,Inference
+28,Serverless Inference API,26,Inference
+29,load_dataset,3,Dataset
+30,DatasetDict,29,Dataset
+31,dataset map,29,Dataset
+32,dataset filter,29,Dataset
+33,evaluate library,1,Evaluation
+34,metric,33,Evaluation
+35,EvaluationSuite,33,Evaluation
+36,DiffusionPipeline,1,Diffusers
+37,StableDiffusionPipeline,36,Diffusers
+38,Accelerator,1,Accelerate
+39,distributed training,38|10,Accelerate
+40,GGUF format,2,Quantization
+41,GGML format,40,Quantization
+42,llama.cpp integration,41,Quantization
+43,bitsandbytes,17,Quantization
+44,Gradio Blocks,4,Demo
+45,Gradio Interface,44,Demo
+46,ChatInterface,44,Demo

ckg_ai_platforms-0.1.0/src/ckg_ai_platforms/domains/langchain.csv

@@ -0,0 +1,49 @@
+ConceptID,ConceptLabel,Dependencies,TaxonomyID
+1,LangChain Framework,,Platform
+2,LangChain Core,1,Library
+3,LangChain Community,1,Library
+4,LangChain CLI,1,Tool
+5,LangSmith,1,Observability
+6,LangChain Hub,1,Registry
+7,ChatPromptTemplate,2,Prompt
+8,PromptTemplate,2,Prompt
+9,MessagesPlaceholder,7,Prompt
+10,BaseLLM,2,Model
+11,ChatOpenAI,10,Model
+12,ChatAnthropic,10,Model
+13,Ollama,10,Model
+14,LCEL RunnableSequence,2,Orchestration
+15,RunnableParallel,14,Orchestration
+16,RunnablePassthrough,14,Orchestration
+17,RunnableLambda,14,Orchestration
+18,LLMChain,10|8,Chain
+19,SequentialChain,18,Chain
+20,RouterChain,18,Chain
+21,AgentExecutor,2,Agent
+22,ReAct Agent,21,Agent
+23,Tool-Calling Agent,21,Agent
+24,OpenAI Functions Agent,21,Agent
+25,BaseTool,2,Tool
+26,StructuredTool,25,Tool
+27,tool decorator,25,Tool
+28,Toolkit,25,Tool
+29,ConversationBufferMemory,2,Memory
+30,ConversationSummaryMemory,29|10,Memory
+31,VectorStoreMemory,29,Memory
+32,BaseRetriever,2,Retrieval
+33,VectorStoreRetriever,32,Retrieval
+34,MultiQueryRetriever,32|10,Retrieval
+35,ContextualCompressionRetriever,32,Retrieval
+36,ParentDocumentRetriever,32,Retrieval
+37,EnsembleRetriever,32,Retrieval
+38,BaseDocumentLoader,2,Loader
+39,WebBaseLoader,38,Loader
+40,PyPDFLoader,38,Loader
+41,UnstructuredLoader,38,Loader
+42,RecursiveCharacterTextSplitter,2,Splitter
+43,TokenTextSplitter,42,Splitter
+44,OpenAIEmbeddings,2,Embedding
+45,HuggingFaceEmbeddings,2,Embedding
+46,FAISS,33,VectorStore
+47,Chroma,33,VectorStore
+48,CallbackManager,5,Observability

ckg_ai_platforms-0.1.0/src/ckg_ai_platforms/domains/langgraph.csv

@@ -0,0 +1,43 @@
+ConceptID,ConceptLabel,Dependencies,TaxonomyID
+1,LangGraph Framework,,Platform
+2,StateGraph,1,Core
+3,CompiledGraph,2,Core
+4,Node,2,Core
+5,Edge,2,Core
+6,Conditional Edge,5,Core
+7,AgentState,2,State
+8,TypedDict,7,State
+9,Annotated Reducer,7|8,State
+10,MessagesState,7,State
+11,START,2,Control
+12,END,2,Control
+13,Branch,6|2,Control
+14,Router Node,4|13,Control
+15,MemorySaver,2,Persistence
+16,SqliteSaver,15,Persistence
+17,PostgresSaver,15,Persistence
+18,AsyncSqliteSaver,15,Persistence
+19,Checkpointer,15,Persistence
+20,thread_id,19,Persistence
+21,interrupt,2,HumanInLoop
+22,Command,21,HumanInLoop
+23,HumanMessage,21,HumanInLoop
+24,approve/reject pattern,21|22,HumanInLoop
+25,stream,3,Streaming
+26,astream,25,Streaming
+27,stream_events,25,Streaming
+28,stream_mode,25,Streaming
+29,create_react_agent,1,Prebuilt
+30,ToolNode,4|29,Prebuilt
+31,tools_condition,6|29,Prebuilt
+32,Supervisor Agent,4,MultiAgent
+33,Worker Agent,4|32,MultiAgent
+34,Handoff,5|32,MultiAgent
+35,Subgraph,2|3,MultiAgent
+36,add_node subgraph,35|2,MultiAgent
+37,LangSmith Trace,3,Observability
+38,LangSmith Evaluate,37,Observability
+39,LangSmith Dataset,38,Observability
+40,recursion_limit,3,Config
+41,config_schema,3,Config
+42,RunnableConfig,41,Config

ckg_ai_platforms-0.1.0/src/ckg_ai_platforms/domains/metadata.json

@@ -0,0 +1,37 @@
+{
+  "langchain": {
+    "nodes": 48,
+    "edges": 71,
+    "description": "LangChain ecosystem \u2014 chains, agents, memory, retrieval, LCEL, tools, and LangSmith"
+  },
+  "langgraph": {
+    "nodes": 42,
+    "edges": 58,
+    "description": "LangGraph agent framework \u2014 stateful graphs, persistence, human-in-the-loop, multi-agent"
+  },
+  "aws-bedrock": {
+    "nodes": 46,
+    "edges": 67,
+    "description": "AWS Bedrock enterprise AI \u2014 foundation models, agents, knowledge bases, guardrails, fine-tuning"
+  },
+  "anthropic-sdk": {
+    "nodes": 44,
+    "edges": 62,
+    "description": "Anthropic Claude API and SDK \u2014 messages, tools, extended thinking, prompt caching, MCP, batching"
+  },
+  "huggingface": {
+    "nodes": 46,
+    "edges": 65,
+    "description": "HuggingFace ecosystem \u2014 Hub, Transformers, PEFT/LoRA, TRL, Inference, Datasets, Gradio"
+  },
+  "openai-api": {
+    "nodes": 43,
+    "edges": 59,
+    "description": "OpenAI API ecosystem \u2014 chat completions, tools, structured outputs, assistants, fine-tuning, batch"
+  },
+  "vertex-ai": {
+    "nodes": 45,
+    "edges": 63,
+    "description": "Google Vertex AI \u2014 Gemini, generative AI, Agent Builder, Vector Search, pipelines, MLOps"
+  }
+}

ckg_ai_platforms-0.1.0/src/ckg_ai_platforms/domains/openai-api.csv

@@ -0,0 +1,44 @@
+ConceptID,ConceptLabel,Dependencies,TaxonomyID
+1,OpenAI API,,Platform
+2,Chat Completions API,1,Core
+3,system role,2,Message
+4,user role,2,Message
+5,assistant role,2,Message
+6,content array,2,Message
+7,function calling,2,Tool
+8,tool_choice,7,Tool
+9,parallel_tool_calls,7,Tool
+10,tool definition JSON,7,Tool
+11,Structured Outputs,2,Output
+12,response_format,11,Output
+13,JSON schema strict,11|12,Output
+14,Assistants API,1,Assistant
+15,Assistant object,14,Assistant
+16,Thread,14,Assistant
+17,Run,14|15|16,Assistant
+18,Message object,16,Assistant
+19,File object,14,Assistant
+20,Code Interpreter,14,Assistant
+21,File Search,14,Assistant
+22,Embeddings API,1,Embedding
+23,text-embedding-3-small,22,Embedding
+24,text-embedding-3-large,22,Embedding
+25,Fine-tuning API,1,Training
+26,fine_tuning.job,25,Training
+27,training_file JSONL,25,Training
+28,validation_file,25,Training
+29,Moderation API,1,Safety
+30,moderation category,29,Safety
+31,moderation score,29,Safety
+32,Images API,1,Multimodal
+33,DALL-E 3,32,Multimodal
+34,image generation,32,Multimodal
+35,image editing,32,Multimodal
+36,Audio API,1,Multimodal
+37,Whisper ASR,36,Multimodal
+38,TTS,36,Multimodal
+39,Batch API,1,Efficiency
+40,batch job,39,Efficiency
+41,Streaming,2,Efficiency
+42,GPT-4o,1,Model
+43,o3 reasoning model,1,Model

ckg_ai_platforms-0.1.0/src/ckg_ai_platforms/domains/vertex-ai.csv

@@ -0,0 +1,46 @@
+ConceptID,ConceptLabel,Dependencies,TaxonomyID
+1,Vertex AI Platform,,Platform
+2,Gemini 2.0 Flash,1,Model
+3,Gemini 2.0 Pro,1,Model
+4,Gemini Ultra,1,Model
+5,Imagen 3,1,Model
+6,Text Embedding Gecko,1,Model
+7,GenerativeModel,2|3,SDK
+8,GenerationConfig,7,SDK
+9,SafetySettings,7,SDK
+10,Content,7,SDK
+11,Vertex AI Studio,1,Studio
+12,Prompt Design,11,Studio
+13,Grounding Google Search,7,Grounding
+14,Grounding Vertex AI Search,7,Grounding
+15,Agent Builder,1,Agent
+16,Dialogflow CX,15,Agent
+17,Search App,15,Agent
+18,Conversation App,15,Agent
+19,Data Store,15|17,Agent
+20,Vector Search,1,Search
+21,Matching Engine Index,20,Search
+22,IndexEndpoint,21,Search
+23,ANN approximate nearest neighbor,20,Search
+24,Kubeflow Pipeline,1,Pipeline
+25,pipeline component,24,Pipeline
+26,pipeline artifact,24,Pipeline
+27,pipeline DAG,24,Pipeline
+28,CustomJob,1,Training
+29,HyperparameterTuning,28,Training
+30,Distributed Training,28,Training
+31,Model Registry,1,Registry
+32,Model Version,31,Registry
+33,Model Evaluation,31|32,Registry
+34,Model Champion Challenger,33,Registry
+35,Feature Store,1,FeatureStore
+36,EntityType,35,FeatureStore
+37,Feature,35|36,FeatureStore
+38,Online Serving,35,FeatureStore
+39,Batch Prediction,1,Inference
+40,Prediction Endpoint,1,Inference
+41,AutoML,1,AutoML
+42,AutoML Tables,41,AutoML
+43,IAM Vertex,1,Security
+44,VPC Service Controls,43,Security
+45,CMEK,43,Security

ckg_ai_platforms-0.1.0/src/ckg_ai_platforms/graph.py

@@ -0,0 +1,106 @@
+import csv
+import json
+from collections import defaultdict, deque
+from pathlib import Path
+
+DOMAINS_DIR = Path(__file__).parent / "domains"
+
+EDGE_TYPES = {"REQUIRES", "ENABLES", "RELATES_TO", "IMPLEMENTS"}
+
+
+def _parse_dep(token: str) -> tuple[str, str, float | None]:
+    """Split 'ID:EDGETYPE:CONFIDENCE' token.
+    EDGETYPE defaults to REQUIRES. CONFIDENCE is float 0-1 or None if unreviewed.
+    """
+    parts = token.split(":", 2)
+    etype = parts[1] if len(parts) >= 2 and parts[1] in EDGE_TYPES else "REQUIRES"
+    confidence = None
+    if len(parts) == 3 and parts[2]:
+        try:
+            confidence = float(parts[2])
+        except ValueError:
+            pass
+    return parts[0], etype, confidence
+
+
+def load_domain_meta(domain: str) -> dict:
+    """Return domain-level provenance: source_url, build_date."""
+    meta_path = DOMAINS_DIR / "metadata.json"
+    if not meta_path.exists():
+        return {}
+    with open(meta_path) as f:
+        data = json.load(f)
+    return data.get(domain, {})
+
+
+def available_domains() -> list[str]:
+    return sorted(p.stem for p in DOMAINS_DIR.glob("*.csv"))
+
+
+def load_graph(domain: str):
+    csv_path = DOMAINS_DIR / f"{domain}.csv"
+    if not csv_path.exists():
+        raise ValueError(f"Domain '{domain}' not found. Run list_domains() to see available domains.")
+
+    id_to_label, label_to_id, prerequisites, dependents, taxonomy = {}, {}, defaultdict(list), defaultdict(list), {}
+    with open(csv_path) as f:
+        for row in csv.DictReader(f):
+            cid = row["ConceptID"]
+            label = row["ConceptLabel"].strip()
+            raw = [t.strip() for t in row["Dependencies"].split("|") if t.strip()]
+            deps = [_parse_dep(t) for t in raw]  # [(dep_id, etype, confidence), ...]
+            id_to_label[cid] = label
+            label_to_id[label.lower()] = cid
+            taxonomy[cid] = row.get("TaxonomyID", "").strip()
+            prerequisites[cid] = deps
+            for dep_id, etype, confidence in deps:
+                dependents[dep_id].append((cid, etype, confidence))
+
+    return id_to_label, label_to_id, prerequisites, dependents, taxonomy
+
+
+def find_concept(label_to_id: dict, query: str) -> str | None:
+    q = query.lower().strip()
+    if q in label_to_id:
+        return label_to_id[q]
+    for label, cid in label_to_id.items():
+        if q in label:
+            return cid
+    return None
+
+
+def bfs_subgraph(start_id: str, adj: dict, id_to_label: dict, max_depth: int) -> list[dict]:
+    """DFS traversal so parent-child relationships indent correctly in text output."""
+    visited: set = set()
+    results: list[dict] = []
+
+    def _dfs(cid: str, depth: int, etype, conf):
+        if cid in visited or depth > max_depth:
+            return
+        visited.add(cid)
+        results.append({
+            "concept": id_to_label.get(cid, cid),
+            "concept_id": cid,
+            "edge_type": etype,
+            "confidence": conf,
+            "depth": depth,
+        })
+        for n, et, c in adj.get(cid, []):
+            _dfs(n, depth + 1, et, c)
+
+    _dfs(start_id, 0, None, None)
+    return results
+
+
+def prerequisite_chain(start_id: str, prerequisites: dict, id_to_label: dict) -> list[str]:
+    visited, queue, chain = set(), deque([start_id]), []
+    while queue:
+        cid = queue.popleft()
+        if cid in visited:
+            continue
+        visited.add(cid)
+        chain.append(id_to_label.get(cid, cid))
+        for dep_id, _etype, _conf in prerequisites.get(cid, []):
+            if dep_id not in visited:
+                queue.append(dep_id)
+    return chain

ckg_ai_platforms-0.1.0/src/ckg_ai_platforms/server.py

@@ -0,0 +1,211 @@
+"""
+ckg-ai-platforms — AI developer platforms as traversable knowledge graphs.
+
+7 domains · 314 nodes · MCP-native · 11x fewer tokens than RAG
+
+Domains: LangChain · LangGraph · OpenAI API · Anthropic SDK · HuggingFace ·
+         AWS Bedrock · Vertex AI
+
+Edge types: REQUIRES · ENABLES · RELATES_TO · IMPLEMENTS
+
+Usage:
+    uvx ckg-ai-platforms                 # run as MCP server
+    python -m ckg_ai_platforms           # same
+
+Claude Desktop config:
+    {
+      "mcpServers": {
+        "ai-platforms": {
+          "command": "uvx",
+          "args": ["ckg-ai-platforms"]
+        }
+      }
+    }
+"""
+from __future__ import annotations
+from mcp.server.fastmcp import FastMCP
+from .graph import available_domains, load_graph, load_domain_meta, find_concept, bfs_subgraph, prerequisite_chain
+
+_DOMAIN_HINTS: dict[str, list[str]] = {
+    "langchain":     ["langchain", "lcel", "chain", "langsmith", "runnable", "agent executor"],
+    "langgraph":     ["langgraph", "graph state", "state machine", "state graph", "cycl"],
+    "openai-api":    ["openai", "gpt", "chat completion", "embeddings", "assistants", "dall-e", "whisper"],
+    "anthropic-sdk": ["anthropic", "claude", "messages api", "tool use", "prompt caching", "computer use"],
+    "huggingface":   ["huggingface", "hugging face", "transformers", "hub", "spaces", "datasets", "inference api"],
+    "aws-bedrock":   ["bedrock", "aws bedrock", "amazon bedrock", "converse api", "guardrails", "knowledge base"],
+    "vertex-ai":     ["vertex", "google cloud", "vertex ai", "model garden", "gemini", "palm"],
+}
+
+
+def _detect_domain(question: str) -> str | None:
+    q = question.lower()
+    best, top = None, 0
+    for domain, hints in _DOMAIN_HINTS.items():
+        score = sum(1 for h in hints if h in q)
+        if score > top:
+            top, best = score, domain
+    return best
+
+
+mcp = FastMCP(
+    "ckg-ai-platforms",
+    instructions=(
+        "CKG server for AI developer platforms (LangChain, LangGraph, OpenAI, Anthropic, "
+        "HuggingFace, AWS Bedrock, Vertex AI). "
+        "Call list_domains first, then query_ckg or get_prerequisites. "
+        "Edges are typed: REQUIRES / ENABLES / RELATES_TO / IMPLEMENTS."
+    ),
+)
+
+
+@mcp.tool()
+def ask_platform(question: str, domain: str = "") -> str:
+    """Answer a question about an AI platform by traversing the relevant CKG.
+
+    Auto-detects the platform from the question if domain is not specified.
+
+    Args:
+        question: Your question about an AI platform concept or workflow.
+        domain: Optional — one of: langchain, langgraph, openai-api, anthropic-sdk,
+                huggingface, aws-bedrock, vertex-ai
+    """
+    domains = available_domains()
+    if not domain:
+        domain = _detect_domain(question)
+    if not domain or domain not in domains:
+        return (
+            f"Could not detect platform. Available: {', '.join(domains)}.\n"
+            "Re-call with domain= set explicitly."
+        )
+
+    id_to_label, label_to_id, prerequisites, dependents, taxonomy = load_graph(domain)
+    meta = load_domain_meta(domain)
+
+    cid = find_concept(label_to_id, question)
+    if cid is None:
+        for word in sorted(question.lower().split(), key=len, reverse=True):
+            if len(word) > 3:
+                cid = find_concept(label_to_id, word)
+                if cid:
+                    break
+    if cid is None:
+        top = list(label_to_id.keys())[:8]
+        return (
+            f"No concept matched '{question}' in {domain}.\n"
+            f"Sample concepts: {', '.join(t.title() for t in top)}"
+        )
+
+    concept_label = id_to_label[cid]
+    chain = prerequisite_chain(cid, prerequisites, id_to_label)
+    subgraph = bfs_subgraph(cid, prerequisites, id_to_label, max_depth=3)
+
+    lines = [
+        f"# CKG: {domain} — {concept_label}",
+        f"Taxonomy: {taxonomy.get(cid, 'unknown')}",
+        "",
+        "## Prerequisite chain (upstream dependencies)",
+    ]
+    for label in chain[1:]:
+        lines.append(f"  ← {label}")
+
+    lines += ["", "## What this concept enables / depends on"]
+    for entry in subgraph:
+        if entry["concept"] != concept_label and entry["depth"] == 1:
+            lines.append(f"  {entry['edge_type']}: {entry['concept']}")
+
+    if meta:
+        lines += ["", f"Domain: {meta.get('description', '')}"]
+    lines.append("\nSource: ckg-ai-platforms v0.1.0 · graphifymd.com")
+    return "\n".join(lines)
+
+
+@mcp.tool()
+def list_domains() -> str:
+    """List all available AI platform domains in this package."""
+    domains = available_domains()
+    lines = [f"Available domains ({len(domains)}):"]
+    for d in sorted(domains):
+        lines.append(f"  - {d}")
+    lines.append("\nUse query_ckg(concept, domain) or ask_platform(question).")
+    return "\n".join(lines)
+
+
+@mcp.tool()
+def search_concepts(query: str, domain: str) -> str:
+    """Find concepts in a domain by partial name match.
+
+    Args:
+        query: Substring to search for (case-insensitive).
+        domain: One of the available domains.
+    """
+    _, label_to_id, _, _, taxonomy = load_graph(domain)
+    q = query.lower()
+    matches = [(label, cid) for label, cid in label_to_id.items() if q in label][:20]
+    if not matches:
+        return f"No concepts matching '{query}' in {domain}."
+    return "\n".join(
+        f"  - {label.title()} [{taxonomy.get(cid, '')}]"
+        for label, cid in matches
+    )
+
+
+@mcp.tool()
+def query_ckg(concept: str, domain: str, depth: int = 3) -> str:
+    """Return the dependency subgraph around a concept.
+
+    Args:
+        concept: Concept name (partial match supported).
+        domain: Domain to search in.
+        depth: Upstream hops to include (1-5, default 3).
+    """
+    id_to_label, label_to_id, prerequisites, dependents, taxonomy = load_graph(domain)
+    cid = find_concept(label_to_id, concept)
+    if cid is None:
+        return f"Concept '{concept}' not found in {domain}."
+
+    concept_label = id_to_label[cid]
+    chain = prerequisite_chain(cid, prerequisites, id_to_label)
+    subgraph = bfs_subgraph(cid, dependents, id_to_label, max_depth=min(depth, 5))
+
+    lines = [
+        f"## {concept_label} [{taxonomy.get(cid, '')}] — {domain}",
+        "",
+        "### Prerequisites (what you need to know first)",
+    ]
+    for label in chain[1:]:
+        lines.append(f"  ← {label}")
+
+    lines += ["", "### Builds toward (concepts that depend on this)"]
+    for entry in subgraph:
+        if entry["concept"] != concept_label:
+            lines.append(f"  → {entry['concept']} [{entry['edge_type']}]")
+    return "\n".join(lines)
+
+
+@mcp.tool()
+def get_prerequisites(concept: str, domain: str) -> str:
+    """Return the full upstream prerequisite chain for a concept.
+
+    Args:
+        concept: Concept to trace (partial match supported).
+        domain: Domain to search in.
+    """
+    id_to_label, label_to_id, prerequisites, _, _ = load_graph(domain)
+    cid = find_concept(label_to_id, concept)
+    if cid is None:
+        return f"Concept '{concept}' not found in {domain}."
+
+    chain = prerequisite_chain(cid, prerequisites, id_to_label)
+    if len(chain) <= 1:
+        return f"{id_to_label[cid]} has no prerequisites in {domain}."
+    return f"Prerequisite chain for {id_to_label[cid]}:\n" + "\n".join(
+        f"  {'  ' * i}← {label}" for i, label in enumerate(chain[1:])
+    )
+
+
+def main() -> None:
+    mcp.run()
+
+
+if __name__ == "__main__":
+    main()