The graph

The graph

scripts/graphify_export.py turns the library into a graphify graph at graphify-out/graph.json (git-ignored). The repository's .mcp.json serves it to Claude Code as the akashic-graph MCP server, so questions about the library's contents (what a board is based on, which boards copy a pedal, which boards use a part) can be answered from the graph instead of by reading the database or the JSON export.

The graph is a read-only view. data/library.sqlite stays the source of truth: fix data there (through the scraper and Fixing a parse), then rebuild the graph.

Nodes and edges

Node	Id	Label	Attributes
board	board:<circuit id>	<Vendor> <name>	akashic_id, vendor, based_on, effect, price, status (in stock, out of stock, or delisted with the date), url, controls, description
vendor	vendor:<vendor id>	the vendor's name	id, url, license_note
original	original:<key>	the most common spelling	boards, spellings (up to five)
part	part:<category>:<value>	the normalized value	category (IC, transistor or diode)

Edge	From	To
sold_by	board	vendor
based_on	board	original
uses_part	board	part

Originals merge spelling variants of the same name: case, punctuation, a leading "the", and aliases such as Electro-Harmonix / EHX and Pro Co / ProCo. Parts are the ICs, transistors and diodes from the parsed parts lists, by normalized value; resistors, capacitors and pots are left out. Like the export, the graph holds only metadata, part values and links, never documents or schematics.

Rebuild

graphify is installed as a uv tool (uv tool install graphifyy). Run the script with graphify's own interpreter, which has networkx and graphify:

~/.local/share/uv/tools/graphifyy/bin/python scripts/graphify_export.py

It reads the database read-only, takes a few seconds, and prints the node counts by type, the edge count and the number of communities. Rebuild after a scrape. scripts/refresh.sh rebuilds it every week whether or not the site deploys, and logs "graphify not installed" when the interpreter is missing.

Querying

In Claude Code, with the akashic-graph server running (graphify-mcp graphify-out/graph.json):

• get_node on a board returns its facts in attributes.
• get_neighbors on an original lists every board based on it; on a part, every board using it.
• query_graph is a fuzzy keyword search; keep token_budget small.

From a shell, at the repository root (the commands read graphify-out/graph.json by default):

graphify explain "Aion FX Aerolith"                  # a node and its connections
graphify query "boards based on the Klon" --budget 800
graphify path "ProCo RAT" "LM308" --undirected       # edges point from boards, so search both ways