If your agent supports the Model Context Protocol — Claude Code, Claude Desktop, ChatGPT with MCP support, and others — install the P2 Knowledge Base MCP. Your agent gets native tool calls for KB queries; no local cache to manage, always serves the latest published content.
Once installed, just ask questions naturally:
"How does the P2 ADD instruction work?"
"Configure a Smart Pin for UART."
"Explain P2 COG architecture."
See AI-PROMPT-PATTERNS.md for more example queries.
If your agent doesn't support MCP, use the fetch-script BOOTSTRAP:
https://raw.githubusercontent.com/ironsheep/P2-Knowledge-Base/main/deliverables/ai-reference/BOOTSTRAP.md
Fetch that file to get one-command setup for both Unix/macOS and Windows.
- 1000+ content files covering PASM2, Spin2, architecture, smart pins, OBEX community code
- Key-based access (fetch-script) or natural-language queries (MCP)
- Always-current published index (MCP) or local caching (fetch-script)
.p2kb-cache/fetch-kb-file.sh --search uart # Find keys
.p2kb-cache/fetch-kb-file.sh --categories # Browse categories
.p2kb-cache/fetch-kb-file.sh p2kbPasm2Mov # Fetch content(MCP users don't need these — the agent calls p2kb_find / p2kb_get directly.)