spurvik-mcp-server
v1.0.9
Published
Local stdio MCP server for Spurvik setup snippets, unwrapped-call scans, HMAC receipt explanations, marking-hook notes, live Art. 50(1) URL and batch domain scans, C2PA image manifest & CDN-strip checks, EU AI Act risk-tier mapping, 24-language Art. 50 ba
Readme
spurvik-mcp-server
Local stdio MCP server. Version 1.0.9. License: MIT. Copyright (c) 2026 EyeTea Solutions ApS (CVR 37093378).
spurvik-mcp-server --root <project>
spurvik-mcp --root <project>On initialize, the server reports serverInfo.version 1.0.9. It speaks MCP protocol 2025-06-18, and 2024-11-05 when the client requests 2024-11-05. @modelcontextprotocol/sdk 1.32.1 is bundled. Third-party licence text is in dist/THIRD-PARTY-NOTICES.txt.
From [email protected], the bundled SDK matches [email protected]: an overridden human decision is treated as not approved. Continue only when the status is approved.
Tools: spurvik_setup_snippet, spurvik_find_unwrapped_calls, spurvik_explain_receipt, spurvik_marking_hooks, spurvik_scan_live_url, spurvik_batch_scan_domains, spurvik_check_image_c2pa_url, spurvik_classify_ai_act_risk, spurvik_generate_art50_badge, and spurvik_vendor_questionnaire_answers. Setup snippets cover OpenAI, Anthropic, Google GenAI, the Vercel AI SDK, Mastra, OpenAI Agents, and LangGraph. The tools are read-only. Seven run locally and do not use the network. spurvik_scan_live_url sends one public URL to console.spurvik.com (8-second timeout) and returns offlineFallback: true with the report, sandbox and badge.js links when the service cannot be reached; localhost, private IPs and internal hosts are refused before any request. It also returns statusShieldMarkdown, a live Art. 50 status shield for a project README. spurvik_batch_scan_domains scans 1 to 10 public domains the same way (at most 4 at once) and returns gapsFound plus one row per domain; if any entry is not a public host, the whole batch is refused before any request. spurvik_check_image_c2pa_url is a C2PA JUMBF manifest & CDN-strip checker: it sends one public https image URL (PNG, JPEG or WebP, up to 4 MB, port 443) to console.spurvik.com and returns manifestFound, signatureMatches, trustAnchor, claimGenerator, validationCodes, whether an image optimizer (/_next/image, Cloudinary, imgix, /cdn-cgi/image/, Shopify CDN, Jetpack Photon, Webflow CDN) stripped the manifest, and a visible notice="generated" snippet. It reads only; it never signs.
spurvik_classify_ai_act_risk maps a use case to prohibited_art5, high_risk_annex_iii, limited_transparency_art50 or minimal_risk (engineering mapping aid; not legal advice). spurvik_generate_art50_badge returns a one-line badge.js tag, a Next.js <Script>, and WordPress, Framer and shadcn/v0 assets in the 24 official EU languages. spurvik_vendor_questionnaire_answers (new in 1.0.8, every input optional) returns copy-paste Markdown answers to the five EU AI Act questions in enterprise vendor security questionnaires and DPO RFPs (Art. 50(1)/(2), Art. 12 logging, GDPR Art. 5(1)(c) minimisation and Frankfurt residency, Art. 14 approval gates, retention and Annex IV), matched to the Free, Pro, Team or Done-For-You plan. From 1.0.8, text tool output ends with a ## Next steps block of links, and the risk and badge tools add it as a second text item after the JSON.
Repository: https://github.com/spurvik/spurvik Plugin tree: https://github.com/spurvik/spurvik/tree/main/plugins/spurvik
