Nutrition MCP Review: clear product, free access (79/100)

Nutrition MCP is a free AI-connected nutrition tracking service for conversational logging.

Fact	Current crawl
Domain	nutrition-mcp.com
Product	Log meals, calories, and macros through Claude or ChatGPT
Category	AI-connected calorie and macro tracking
Pricing	Free; $0–$0
Platforms	Web, iOS, Android
Pages crawled	17
Crawl date	2026-10-16

Nutrition MCP is a strong free product with unusually concrete positioning for people who already use Claude, ChatGPT, or another MCP client. The public score is 79/100, supported by high scores for usability (93/100), accessibility (97/100), technical SEO (88/100), and first impressions and positioning (83.3/100).

The product explains its core job clearly: conversational meal logging, with the /tools page naming 48 capabilities and their user-facing jobs. The sampled performance evidence is good for real users but less comfortable in mobile lab conditions: the homepage records p75 LCP of 2.113 s, INP of 113 ms, CLS of 0, and TTFB of 760 ms.

Mobile lab performance is the clearest product opportunity: homepage LCP is 3.903 seconds, while /de/tools measures 3.210 seconds. Writing and editorial QA are stronger overall, but both summaries identify localization as the clearest editorial risk across seven sampled localized pages. The result is a focused product with a clear first-use path and a mobile performance opportunity.

Nutrition MCP makes its core promise clear: people can track nutrition by talking to Claude or ChatGPT instead of opening a separate tracker. The homepage backs that promise with automatic calorie and macro logging, barcode lookup, USDA data, photos, summaries, goals, and CSV export. The strongest competitive distinction appears on /alternatives, where MyFitnessPal, Cronometer, and Lose It! are named and the site says the first three have no official MCP server. The main positioning fix is hierarchy. The H1 says “Track your nutrition by talking to your AI,” while “Calorie & Macro Tracker” appears in the title and metadata. Put that category phrase in the first viewport, then bring the official-MCP-server distinction closer to the promise.

A Claude or ChatGPT user can understand the offer, cost, supported clients, and setup path from the supplied crawl. The site describes conversational meal logging, calorie and macro tracking, barcode and photo input, summaries, goals, and CSV export, and it states that the product is completely free. “How it works,” “Install,” “Tools,” “Examples,” and “FAQ” mirror visitor goals, while the installation flow names OAuth 2.0 with PKCE, Google or email sign-in, and client-specific instructions. The remaining messaging gap is confidence after something goes wrong: the supplied artifacts do not expose a concrete answer for cancellation, recovery, or support. Add a compact trust-and-recovery block beside setup, and pair the procedural guidance with one attributable outcome or sourced usage metric.

Nutrition MCP presents a short, understandable path from discovery to use. “How it works,” “Install,” “Tools,” “Examples,” and “FAQ” are plain-language routes, and “Add to Claude” names the destination directly. The supplied simulated desktop and mobile walkthroughs reached both goals without a failed task, and a skeptical evaluator could follow GitHub, Privacy Policy, Terms of Service, or the alternatives page. The main friction is the conversion boundary: the crawl reports no forms, and the primary action leaves for https://claude.ai/directory/nutrition-mcp rather than opening an on-site signup. Add one sentence beside “Add to Claude” explaining what opens next and whether a separate Nutrition MCP account is needed. Make it the consistent primary label, with “Quick install” explained as the technical alternative.

The sampled public surface has solid structural accessibility foundations. The DOM facts for the homepage, /tools, /alternatives, /es, /de, and /it report header, nav, main, footer, and section landmarks; each sample has a skip link, progressing heading sequences, zero images missing alt text, and zero positive tabindex values. The homepage extraction also reports label-or-ARIA coverage for three controls, although that artifact is not a complete accessible-name computation. The confirmed issue is visual contrast: the homepage PSI accessibility audit scores 97 but fails color-contrast because a background and foreground pair lacks sufficient contrast. The failing node and ratio are not supplied, so use Lighthouse to identify the pair, update the relevant tokens to WCAG AA, and rerun representative pages.

The available evidence supports a good design foundation, with low confidence because the crawl covered 17 of 40 pages and supplied no CSS bytes, rendered DOM measurements, or screenshots. The homepage heading outline contains one H1 followed by coherent H2 and H3 groups, led by “Track your nutrition by talking to your AI.” Layout stability is a strong measured signal: PSI reports CLS of 0.001 on mobile and 0.004 on desktop, with accessibility scores of 97 for both strategies. Both PSI runs still fail the color-contrast audit. Because the failing pairs are not named, the next design task is to identify them in Lighthouse and correct the corresponding tokens. Treat spacing, typography, focus states, tap targets, overflow, and icon consistency as unmeasured until the missing visual artifacts are captured.

Nutrition MCP has a good real-user performance signal and a clear mobile lab opportunity. In the supplied PageSpeed field samples, the homepage records p75 LCP of 2.113 seconds, INP of 113 ms, CLS of 0, and TTFB of 760 ms. The sampled localized tools pages record p75 LCP of 1.916 seconds, INP of 104 ms, CLS of 0, and TTFB of 791 ms. The mobile lab is slower: homepage LCP is 3.903 seconds, while /de/tools measures 3.210 seconds. PageSpeed flags render-blocking requests, estimating 2,880 ms on the homepage and 2,240 ms on /de/tools. Use a mobile waterfall to identify the actual CSS, font, or script blockers before changing the critical path.

The strongest writing pattern is specific and immediate. “Track your nutrition by talking to your AI” is followed by “Connect Claude or ChatGPT, then just say what you ate. Calories and macros, logged automatically.” The /tools page carries that precision into reference content with “48 tools” and names such as log_meal, lookup_barcode, and export_all_data. The main editorial risk is localization workflow: the Spanish page says, “Esta página fue traducida con IA a partir del inglés y no ha sido revisada por una persona,” and equivalent notices appear on the sampled German, Italian, Dutch, Polish, French, and Ukrainian pages. Rendered-text averages also reach 26.5 words per sentence on /tools, 27.3 on /es, and 25.8 on /de/tools. Prioritize native-language review and split long action descriptions. Remove the space before the period in the /alternatives H1.

The supplied crawl shows stable technical behavior rather than a confirmed traffic-loss mechanism. All 17 crawled URLs returned 200 with self-canonicals; the verified sitemap contains 120 URLs; the HTTP homepage redirected once to HTTPS; and raw and rendered content matched across the sample. The main hedgeable risk is mobile lab performance: homepage LCP was 3.903 seconds and sampled tools pages measured 3.003–3.210 seconds, with render-blocking and document-latency findings. Available CrUX field data was FAST, including homepage p75 LCP of 2.113 seconds and /de/tools p75 LCP of 1.916 seconds, so this is an optimization signal rather than evidence of active traffic loss. Connect GSC and analytics before attributing any change to performance, and complete the full crawl before extending the sampled stability conclusion sitewide.

The primary English surface clears a basic human-editor bar: the home page answers what the product is, and /tools follows “48 tools” with named, inspectable capabilities and user-facing descriptions. The clearest QA gap is localization release control. Seven sampled pages—Spanish, German, Italian, Dutch, Polish, French, and Ukrainian—disclose that their AI translations were not reviewed by a person. That disclosure is useful evidence of process status, not a defect in writing quality by itself; the practical fix is a native-language checklist covering the hero, installation, tool names, consent, and legal-adjacent copy. Mechanical follow-up is also due: /tools, /es, and /de/tools average 26.5, 27.3, and 25.8 words per sentence, and the crawler flags the 67-character /de/tools title.

Nutrition MCP's machine-readable orientation is useful for a developer-facing MCP service. The supplied llms.txt file describes the MCP connection URL, supported nutrition workflows, data-source behavior, and import/export concepts. That file should remain, and it should be paired with canonical human-facing pages. Publish a /docs or /getting-started hub, link it from navigation or the footer, and include one end-to-end MCP connection example, tool inputs and examples, limitations, and troubleshooting for connection, logging, and data sources.

The partial crawl shows a healthy technical SEO foundation. All 17 crawled URLs returned 200 and used self-referencing canonicals, while the verified sitemap contains 120 URLs. robots.txt allows the public site, declares the sitemap, and disallows only /api/ and /mcp; the HTTP homepage reaches HTTPS in one hop. Raw and rendered word counts are effectively identical across the 17 sampled pages, so the supplied routes deliver their content without a meaningful JavaScript-only gap. The main measured opportunity is mobile speed: PSI reports LCP of 3.903 seconds on the homepage and about 3.2 seconds on sampled tools pages, moderated by FAST CrUX field values. Structured data, response security headers, screenshots, complete sitemap coverage, and full hreflang reciprocity were not supplied, so capture those artifacts before making sitewide claims.

Nutrition MCP earns 79/100 because it makes conversational nutrition logging easy to understand and keeps the product free. The strongest evidence is practical: usability scores 93/100, accessibility scores 97/100, and the homepage offers a direct “Add to Claude” handoff alongside clear “How it works,” “Install,” “Tools,” “Examples,” and “FAQ” paths.

The clearest improvement areas are public documentation, localization, and mobile lab performance. The 9,293-byte llms.txt file provides a machine-readable orientation for the service, while the supplied lab results are slower than the field results. Documentation should be paired with canonical human-facing pages, and localization needs tighter editorial attention.

This product fits Claude, ChatGPT, and other MCP-client users who want to log nutrition conversationally without a separate tracker app. Its free model lowers the barrier to trying it; documentation will determine how easily less experienced users can continue beyond the first setup step.

This 13-dimension review combines the supplied public score table with evidence from a 17-page crawl completed on 2026-10-16. The inputs include sampled page structure, performance field data, homepage lab reports, product facts, sitemap and rendering checks, and content samples. The overall article score is the rounded persisted score of 79/100.

Two dimensions are excluded from the article sections: 12 · Decision-support surfaces (failed) and 13 · Review-content integrity (not applicable). Google Search Console enrichment was not available because GSC was not connected and access was not granted.

Read How SiteList scores for the full scoring approach and data notes.