🧠 Memory Workflows¶

These prompts demonstrate persistent AI memory using OpenAI vector stores — one of the most powerful patterns in MCP.

Requires OpenAI API key

Set OPENAI_API_KEY in your .env before trying these.

Pattern 1: Build a Knowledge Base¶

Save facts across multiple sessions, then retrieve them semantically.

Step 1 — Fill the memory store:

Please save the following facts to my memory store:
1. "MCP stands for Model Context Protocol, an open standard by Anthropic"
2. "FastMCP is a Python library that simplifies building MCP servers"
3. "MCP supports three primitives: Tools, Prompts, and Resources"
4. "MCP ToolHub uses pyautogui for screenshot capture"
5. "The MCP Inspector is a browser-based debugger at localhost:5173"
6. "Vector stores in OpenAI enable semantic search over stored text"

Confirm each one was saved.

Step 2 — Retrieve by meaning (not keyword):

Search my memory for anything about "debugging MCP"

What do I know about "Python libraries for AI servers"?

Search memory: "how to capture screen"

Pattern 2: Session Notes¶

Use memory as persistent session notes that survive across conversations.

I'm starting a new learning session. Save to memory:
"Session 1 complete: installed MCP ToolHub, tested all tools, 
confirmed screenshot and memory tools working. 
Next: explore prompt templates and resource URIs."

In a future session:

What did I accomplish in my last MCP learning session? Search my memory.

Pattern 3: Personal Knowledge Assistant¶

Save these personal preferences to memory:
- "I prefer Perplexity for web search over OpenAI"
- "My vector store is named MCP_ToolHub_Memories"
- "I always run uv with VIRTUAL_ENV unset"

Then search for "my preferences" to verify they're stored.

Pattern 4: Research → Save → Recall¶

Combine web search with memory for a research workflow:

1. Search the web for: "latest MCP specification updates 2025"
2. Save a summary of the key points to memory
3. Then immediately search memory for "MCP specification" to verify it saved

Pattern 5: Memory + Analysis¶

Search my memory for everything saved about MCP.
Then use the analyze_topic prompt to do a structured analysis of
"Model Context Protocol" based on what you found.

Why vector memory matters

Unlike a notes file (exact text lookup), vector memory enables semantic search — you can ask "what do I know about debugging?" and it finds relevant stored facts even if they don't contain the word "debugging".