Are you the author? Sign in to claim
Read-only remote MCP server for wafergraph.com's semiconductor and AI supply-chain dataset (565 companies, supplier/cust
A remote MCP server exposing wafergraph.com's semiconductor & AI supply-chain dataset — 565 companies across 12 segments, the supplier/customer graph, and a 74-deal M&A corpus — as nine read-only tools any MCP-speaking AI agent can call directly.
No auth, no cost, read-only. Streamable HTTP transport at /mcp. Human landing page at /.
Live: https://wafergraph-mcp.jwpalm99.workers.dev
| Tool | Purpose |
|---|---|
search_companies({query?, segment?, country?}) | Name/description search with segment & country filters. Compact list, capped at 25, with a total match count. |
get_company({id}) | Full allowed profile for one company (by id or name) plus its supplier/customer edges. |
get_segments() | The 12-segment taxonomy (+ subsegments, market_position enum) with live company counts. |
get_supply_chain({id, direction, depth}) | Walk the supplier/customer graph from a focal company up to 2 tiers up ("up"), down ("down"), or "both". |
get_deals({query?, segment?}) | Search the M&A deal corpus. Compact list, capped at 30, with a total match count. |
compare_companies({ids}) | Side-by-side comparison of 2-6 companies on aligned fields, plus their shared and unique supply-chain counterparties. |
get_country_exposure({segment?}) | Geographic concentration: which countries host a segment's companies, ranked by count, with disclosed market-cap coverage. |
find_chokepoints({segment?, limit?}) | Rank chokepoints by downstream dependency weighted by market position. Transparent heuristic, not a proprietary risk model. |
analyze_portfolio_exposure({holdings}) | Map tickers or ids to segment/country exposure and flag suppliers that several holdings share. Not investment advice. |
Every response includes:
data — the payload.attribution — compiled-by/sources block (SEC, Wikidata, Wikipedia, GLEIF) plus a
company_url back to the full sourced profile on wafergraph.com.links — { report, newsletter } pointing back to wafergraph.com and its $99 report.Company records include key_products as of 2026-07-19. That field in wafergraph's own dataset
was originally bulk-drafted with a known fabrication/mis-scope history and was withheld here
until a verified deep-fill landed upstream; it now has (confirmed live) 564/565 companies filled,
with only sk_enpulse empty (defunct, absorbed into SKC 2025-12-23). Every field shipped here —
name, ticker, market_cap_usd_b, segments, one_liner, market_position, key_products,
key_customers/key_suppliers graph edges, deals — is established, trust-checked data. Fields
are still whitelisted (not blacklisted) in src/types.ts (toAllowedCompany), so anything added
upstream later stays excluded by default until deliberately added.
Counters only, never identities. No IPs, no query contents or tool arguments, no user
identifiers, no cookies, no PII. Specifically: how many times each tool was called per day, how
many distinct sessions opened per day, and which client software connected (the clientInfo
name and version your MCP client already sends in the initialize handshake, e.g.
claude-code@2.1.0).
Sessions are not users — one person reconnecting counts several times. Nothing recorded here can identify who you are.
claude mcp add --transport http wafergraph https://wafergraph-mcp.jwpalm99.workers.dev/mcp
Settings → Connectors → Add custom connector → paste:
https://wafergraph-mcp.jwpalm99.workers.dev/mcp
{
"mcpServers": {
"wafergraph": {
"url": "https://wafergraph-mcp.jwpalm99.workers.dev/mcp"
}
}
}
Cloudflare Worker, TypeScript, built on the agents package's McpAgent (Durable-Object-backed
Streamable HTTP MCP server — free on the Workers Free plan; SQLite storage backend). See
CLAUDE.md for the file map and the data-source-mode decision (hybrid live-fetch + one
vendored snapshot, discovered by E2E-testing the upstream endpoints rather than assumed).
npm install
npm run dev # wrangler dev, local
npm run typecheck # tsc --noEmit
npm run deploy # wrangler deploy
See ATTRIBUTION.md for the full source/license breakdown (SEC EDGAR public domain, Wikidata
CC0, Wikipedia CC BY-SA 4.0 — attribution required, GLEIF). Short version: free to use, credit
wafergraph.com, not financial advice.
This is an independent project built against wafergraph.com's public dataset; it is not an official wafergraph product.
Run Claude Code as an MCP server so any agent can delegate coding tasks to it
Browser automation using accessibility snapshots instead of screenshots
Google's universal MCP server supporting PostgreSQL, MySQL, MongoDB, Redis, and 10+ databases
Official GitHub integration for repos, issues, PRs, and CI/CD workflows