Torque Index for AI assistants

MCP server

Connect Claude, ChatGPT, Cursor or any Model Context Protocol client and it can look up a truck VIN, find a part number and read every page of this reference: 27 engine platforms, 1,191 part numbers. Free, no key, read-only.

Server address

https://torqueindex.com/mcp

Transport: Streamable HTTP (JSON answers, stateless). Registry name: com.torqueindex/truck-reference.

Connect

Claude (claude.ai or Desktop): Settings → Connectors → Add custom connector, paste the address above. Claude Code:

claude mcp add --transport http torqueindex https://torqueindex.com/mcp

Cursor, VS Code and other clients (mcp.json):

{
  "mcpServers": {
    "torqueindex": { "type": "http", "url": "https://torqueindex.com/mcp" }
  }
}

Tools

lookup_vin(vin)

Decodes a 17-character VIN through NHTSA vPIC and returns the matching engine platform, vehicle configuration pages and the recalls and bulletins that apply. The same answer as the VIN lookup and its JSON API.

find_part_number(number)

Which part a Ford, Mopar, GM or aftermarket number is, which engine it fits and what it replaced. The same search as the part number lookup.

get_page(path)

The markdown version of any page here, e.g. /platforms/cummins-6_7/, a failure mode, a part, a recalls page or a blog article.

list_platforms()

Every engine platform with its page, recalls page and part count.

get_failure_modes(platform)

Every documented failure mode of one engine: name, provenance tier, how common, first symptom. The page has root cause, diagnosis and fixes.

get_recalls(platform)

NHTSA safety recalls on record for one engine: campaign number, date, component, model years, summary. The recalls page adds investigations and bulletins.

get_repair_cost(platform)

The parts budget of every priced failure on one engine, Dave's prices exact, other parts approximate - parts only, no labor. Same numbers as /repair-costs/.

get_service_specs(platform)

Owner's-manual cards: oil capacity, grade and the maker's oil spec, filter numbers, coolant - per manual year. Same data as /service-specs/.

get_symptoms(symptom, platform?)

"My truck hard-starts" → the documented failure modes that show that sign, per engine. Lists causes, does not diagnose. Same index as /symptoms/.

search(query)

Finds concept pages by words: hubs, failure modes, recipes, symptom pages, blog guides. For parts use find_part_number, for a specific truck lookup_vin.

Try it without a client

curl -s https://torqueindex.com/mcp -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"find_part_number","arguments":{"number":"F81Z-9N184-AA"}}}'

Limits and honesty

Read-only: the tools only read this reference and NHTSA's VIN decoder. Nothing is stored or logged.
Every fact page states a provenance tier (verified / aggregated / contested / data-gap); the server asks the assistant to state the tier and link how we source facts.
Fitment is year × model × engine: cab and wheelbase are not modeled.
The site allows about 8 requests per 10 seconds from one address.