Tracely is a hosted MCP server that gives your AI coding assistant real automotive-compliance judgment: ASIL determination, ASPICE 4.0 work products, requirement / architecture / design review, and the anti-patterns that fail assessments — retrieved on demand instead of guessed. Two lines of config and your assistant stops hallucinating base-practice IDs.
Tracely is the self-serve front door to automotive compliance for AI-assisted engineering. It plugs into Claude Code, Cursor, Windsurf, or GitHub Copilot as an MCP server and hands your assistant a set of tools that return grounded ASPICE 4.0 and ISO 26262 judgment. It is the fast way in; producing assessment-ready evidence end to end from your codebase is what Vera does.
ASPICE process IDs, base practices, the ASIL matrix, decomposition rules, work-product definitions — these are exactly the facts a general-purpose LLM gets subtly wrong. ASPICE 3.1 and 4.0 differ; the S/E/C-to-ASIL matrix has edge cases; "which work products does SWE.3 actually require" is not something to guess in front of an assessor. Tracely's defining trait is that it never guesses. Every process detail, ISO clause, ASIL value, and work-product requirement is retrieved from the server and then interpreted in your context. The tools are the source of truth; your assistant is the interface.
determine_asil classifies a hazardous event from S/E/C per the ISO 26262-3 matrix; check_asil_decomposition validates decomposition and independence per ISO 26262-9. The edge cases are handled for you.
get_work_products, assess_process_compliance, and check_traceability gap-analyze your evidence against ASPICE 4.0 base practices and verify bidirectional V-model trace chains.
review_requirement, draft_requirement, and suggest_verification_method check requirements for ambiguity, testability, and atomicity — and pick the right verification method by ASIL.
review_architecture, review_interface_specification, evaluate_dynamic_behavior, review_detailed_design, check_design_quality_metrics, and review_resource_allocation apply the assessment lenses directly to your descriptions.
get_anti_patterns, consult_lukas, and get_consulting_framework surface what not to do — the documented failure modes that repeat across OEMs and Tier-1s — plus a structured consulting perspective from a practicing ASPICE change agent.
Seventeen tools in total, plus browsable resources covering every ASPICE 4.0 process area (SWE / SYS / MAN / SUP), ISO 26262 parts 1–12, capability-level definitions, and ASPICE↔ISO 26262 mappings. Structured inputs, grounded outputs, every finding cited to a specific base practice or clause.
ask_live_<…> to authenticate. Don't have one yet? Ask for a trial key →.mcp.json (or ~/.claude.json for a global install). Note the type is http — Claude Code silently loads zero tools if you use url.{
"mcpServers": {
"tracely": {
"type": "http",
"url": "https://tracely-mcp.autospec.workers.dev/mcp",
"headers": {
"Authorization": "Bearer $TRACELY_API_KEY"
}
}
}
}$TRACELY_API_KEY reference resolves (or paste the raw key inline):
export TRACELY_API_KEY="ask_live_your_key_here"tracely-aspice skill plus an installer). This teaches your assistant when to reach for each tool and which anti-patterns to avoid — so it behaves like an ASPICE engineer, not just a repo with a plugin.Review my software architecture for ASPICE SWE.2 or Determine the ASIL for loss of steering assist at highway speed. The right tools are called automatically.Prefer to wire it up automatically? The download also includes install_tracely.py, which merges the tracely entry into ~/.claude.json without touching the rest of your config: python install_tracely.py --key ask_live_<…>, then --status to verify.
| Tracely | Vera | |
|---|---|---|
| What it is | Self-serve MCP that makes your AI assistant ASPICE-literate | The AI ASPICE engineer that builds assessment evidence from your code |
| You bring | Your repo + your coding tool | Your project; Vera runs the compliance production |
| Output | Grounded judgment on demand — ASIL, work products, review verdicts | Assessment-ready work products, each with a confidence score and audit trail |
| Pricing | Metered access — the front door | Outcome-priced engagement |
“Tracely doesn't replace your assessor or your process ownership. It removes the part teams actually fail on: guessing the standard instead of retrieving it.”The practitioner view
A hosted MCP (Model Context Protocol) server that makes any Claude Code or Copilot repository automotive-compliance-literate. Your assistant gains tools for ASIL determination, ASPICE 4.0 work products, requirement / architecture / design review, traceability checks, and the anti-patterns that fail assessments — retrieved on demand rather than guessed.
Add a single mcpServers.tracely entry to your project's .mcp.json pointing at https://tracely-mcp.autospec.workers.dev/mcp with an Authorization: Bearer header carrying your API key, then relaunch your tool. Optionally drop the downloadable Tracely skill into your repo so the assistant knows when to reach for each tool.
Any MCP-capable coding assistant — Claude Code, Cursor, Windsurf, and GitHub Copilot. It speaks Streamable HTTP over a single endpoint.
No. Tracely is the front door — it makes your existing repo and assistant fluent in ASPICE and ISO 26262 so you stop guessing base-practice IDs and ASIL values. Producing assessment-ready evidence end to end from your code is what Vera does; Tracely is the fast, self-serve way in.
No. Tracely is a tool. It exposes derived judgment about ASPICE 4.0 (the VDA process-assessment model) and ISO 26262 (functional safety) so your assistant can cite the right process, clause, or ASIL input instead of hallucinating one. It is not a standard and not a legal requirement itself.
Grounded in the standard, honest about the theater: What is ASPICE · Automating ASPICE with AI agents · Agile ASPICE · Best ASPICE tools · The Blueprint (all process areas)