MCP Hive Human-Dependent

Name: Agent Native Offer Audit — MCP Hive
Item: MCP Hive
Rating: 42
Author: Agent Native Offers

AUDIE Score: 42/100 · Audited 2026-04-23 · Website: https://mcp-hive.com · Machine-readable: JSON

Pillar Scores

P1 Signal Architecture — 15/25

P2 Clarity Stack — 11/25

P3 Trust Envelope — 5/20

P4 Velocity Triggers — 5/10

P5 Gravity Design — 6/20

Executive Summary

MCP Hive scores 42/100 (Human-Dependent), reflecting a platform with strong structural intent — subscription access to a curated MCP catalog, excellent schema.org pricing markup, and standard MCP protocol as the connection layer — but significant agent-readiness gaps stemming from its pre-launch status and missing agent-native infrastructure. The platform's biggest single weakness is the complete absence of a machine-readable identity layer (llms.txt), which forces all agent discovery through human-written documentation. Priority actions are both low-effort and high-impact: publishing /llms.txt and a usage API before the May 11, 2026 launch would immediately move MCP Hive from Human-Dependent to the Emerging tier, and the platform's network-effect model gives it genuine long-term potential to reach Agent-Ready if the provider ecosystem grows as intended.

Strongest Signals

Best-in-class schema.org pricing markup The pricing page's SoftwareApplication + Offer JSON-LD implementation is among the most complete seen in this audit series — all three tiers with prices, availability, return policy, and validity dates in structured format.
Standard MCP protocol as the access layer By using MCP as the connection standard, MCP Hive is natively agent-compatible. Any agent that speaks MCP can access the Hive without custom integration work.
30-day free trial with no credit card A genuine friction-free evaluation signal that agents (and their human operators) can act on without financial commitment.

Critical Gaps

No llms.txt or agent identity layer Agents cannot programmatically discover what MCP Hive offers, how it works, or how to integrate — all discovery requires human-readable documentation.
Per-tool pricing opacity Subscription buys access, but what individual MCP tools cost to invoke (especially "priced MCPs") is not centrally machine-readable. Agents cannot compute total operational cost from a single source.
Pre-launch status creates near-total trust deficit No verifiable performance data, no live provider track record, Professional and Enterprise in PreOrder — agents cannot evaluate the platform on operational evidence.
No substitution or fallback rules If a specific MCP server within the Hive goes down, there is no documented fallback policy. For mission-critical agent workflows, this is a production blocker.

Priority Actions

Publish /llms.txt with gateway description, available server categories, and integration instructions — +4 pts · P1 · Effort: Low
Add a public quota/usage API endpoint for subscribers — +3 pts · P3 · Effort: Medium
Publish per-MCP pricing in a centralized machine-readable catalog — +3 pts · P2 · Effort: Medium
Define and publish substitution and fallback policies — +3 pts · P2 · Effort: Low
Launch a public status page on or before launch day (May 11, 2026) — +2 pts · P3 · Effort: Low

All 20 Criteria

P1-A Structured Data — 4/5
Rich schema.org implementation: Organization schema on every page, plus SoftwareApplication schema on /pricing with a complete Offer array for all three tiers (Starter, Professional, Enterprise) including price, priceCurrency, priceValidUntil, availability, and MerchantReturnPolicy. Missing AggregateRating for full score. Among the better structured data implementations seen in this audit series.

P1-B Machine-Readable Pricing — 5/5
All three subscription tiers expressed as schema.org/Offer objects: Starter $19.99/mo (8k calls), Professional $39.99/mo (16k calls + $20 MCP credit), Enterprise $249.99/mo (pay-as-you-go + custom caps). Pricing also displayed in clean HTML text. A machine agent can parse all tier options from structured JSON-LD without page scraping.

P1-C llms.txt / Agent Layer — 1/5
No /llms.txt found. No agent-specific identity layer. Documentation at /docs is human-written prose. robots.txt explicitly disallows /api/ paths, limiting programmatic discovery. No hidden agent-readable metadata found.

P1-D API / MCP Availability — 3/5
MCP Hive operates as a gateway to MCP servers — agents connect via standard MCP protocol using the mcpProxy npm package (@mcp-hive-utils/proxy). Gateway and Server modes supported per docs. No public REST API or OpenAPI specification for the MCP Hive gateway itself. Standard protocol compliance is positive; lack of a documented machine-readable API spec for the gateway is a gap.

P1-E Discoverability (GEO) — 2/5
robots.txt allows all public pages; sitemap.xml present. Standard SEO metadata on all pages. No evidence of AI retrieval optimization, no agent-specific content discovery mechanisms beyond standard web indexing.

P2-A Offer Completeness — 3/5
Subscription tier pricing is clear and machine-readable (via schema.org/Offer). However, the actual cost of using individual MCP servers varies by provider — this per-tool pricing is not centrally machine-readable and requires navigating individual provider pages. The core offer (what does an agent actually pay per tool call?) is not expressible from a single URL.

P2-B Scope & Limits — 3/5
Monthly call limits stated per tier (8k, 16k calls). Enterprise includes custom spending caps. $20/month priced-MCP credit on Pro/Enterprise documented. Missing: no explicit documentation of what happens at quota exhaustion (blocking vs. overage billing?), and per-tool rate limits are not specified.

P2-C Substitution Rules — 1/5
No guidance found on what occurs when a specific MCP server within the Hive goes offline, is deprecated, or returns errors. No stated fallback routing, SLA on server availability, or provider replacement policy. This is a meaningful gap for production agent deployments.

P2-D Conditional Logic — 2/5
Subscription tiers are disclosed. However, Professional and Enterprise plans are listed as "PreOrder" availability in schema.org (confirming pre-launch status as of audit date). Conditions around overage pricing, provider-specific pricing rules, and behavior at plan limits are not clearly disclosed in machine-parseable format.

P2-E Semantic Precision — 2/5
Marketing copy uses imprecise terms: "quality-assured," "verified quality stats," "commercial-grade MCP servers." The promised metrics (accuracy, latency, coverage) are mentioned but not defined, benchmarked, or accessible in structured form. Subscription descriptions are precise (call counts, credit amounts) but capability claims lack operational definitions.

P3-A Verifiable Performance — 1/5
Claims "verified quality stats" (accuracy, latency, coverage) but these are not publicly accessible or third-party verified. No status page found. Platform is pre-launch (May 11, 2026), meaning no historical performance record. Founding Provider program currently recruiting first 100 providers — product is not yet live.

P3-B Scoped Permissions — 2/5
Monthly call limits per subscription tier create spending boundaries. Enterprise tier offers custom spending caps — this is an agent-relevant feature. No per-provider or per-tool permission scoping documented. No time-bounded or action-bounded access controls.

P3-C Audit Trail — 1/5
robots.txt disallows /usage/ paths, suggesting a usage dashboard exists behind authentication. No machine-accessible audit log API documented. Usage history appears to be human-facing only.

P3-D Behavioral Consistency — 1/5
Terms of Service and Privacy Policy pages exist. No version-controlled terms, no changelog, no stated change notice policy. Pre-launch status means no behavioral track record. Provider application process (8+ days per DEV.to account) suggests manual, inconsistent provider onboarding.

P4-A Friction-Free Activation — 3/5
Self-serve signup with 30-day free trial, no credit card required. However, activation requires: (1) creating an account, (2) obtaining API credentials from dashboard, (3) installing mcpProxy npm package, (4) configuring it with credentials. Multi-step setup with npm dependency, and Professional/Enterprise plans are in PreOrder status, limiting immediate access.

P4-B Agent Decision Signals — 2/5
30-day free trial is an explicit signal for when to try the platform. Tiered pricing with stated call limits helps agents model monthly costs. No programmatic trial eligibility check, no agent-readable signals for when to upgrade tiers, no free exploration without signup.

P5-A Integration Depth — 2/5
mcpProxy npm dependency creates mild technical switching cost. Single-gateway model creates workflow dependency. The provider ecosystem is the core value proposition — if MCP Hive successfully recruits 100+ quality providers, this becomes the primary agent tool catalog, creating significant network-effect gravity. Currently pre-launch with unconfirmed provider count.

P5-B Agent Memory Layer — 1/5
No memory or personalization layer found. Dashboard provides usage history (human-facing). No API to retrieve an agent's historical usage patterns, preferences, or per-provider performance data.

P5-C Programmatic Renewal — 1/5
Subscription implies auto-renewal for human subscribers, but no agent-accessible renewal API or subscription management endpoint documented. An agent cannot programmatically check remaining call quota, upgrade tiers, or manage its own subscription.

P5-D Compounding Value — 2/5
Founding Provider program creates a growing catalog: more providers = more tool options per subscription = increasing value over time. This is a genuine compounding value model. However, no agent-readable signal of catalog growth rate, new provider additions, or quality improvement over time.

Rubric v1 (April 2026). Scores reflect the company's state on the audit date and may have improved since.