MCPMarket Human-Dependent

Name: Agent Native Offer Audit — MCPMarket
Item: MCPMarket
Rating: 58
Author: Agent Native Offers

AUDIE Score: 58/100 · Audited 2026-04-10 · Website: https://mcpmarket.com · Machine-readable: JSON

Pillar Scores

P1 Signal Architecture — 13/25

P2 Clarity Stack — 10/25

P3 Trust Envelope — 10/20

P4 Velocity Triggers — 10/10

P5 Gravity Design — 15/20

Criterion-level scores sum to 56; the published pillar totals (authoritative) sum to 58. Discrepancy traces to the original audit document and is preserved for transparency.

Executive Summary

MCPMarket lands in the **Human-Dependent** tier (58/100) — but the score is deceptive. It achieves a perfect Velocity score (10/10), the only platform audited to date to do so, because it requires zero friction for discovery and provides complete decision-making signals upfront. The structural moat is also real: strong network effects and deep integration gravity give MCPMarket lasting platform potential. The gap is almost entirely in Signal Architecture and Clarity — the platform that exists to help AI agents discover tools has no /llms.txt, no schema.org markup, and no programmatic API of its own. Publishing /llms.txt (30-minute fix) and adding SoftwareApplication schema markup would immediately lift the score by 9+ points. A lightweight directory API would push MCPMarket firmly into Agent-Ready territory and cement its position as the canonical MCP discovery layer for autonomous agents.

Strongest Signals

Zero-friction activation (P4-A: 5/5) MCPMarket is the purest zero-friction offer in any audit to date. No signup, no gate, no human handoff. Agents and developers can discover and deploy MCP servers in a single session.
Agent decision signals (P4-B: 5/5) Install commands, capability descriptions, required permissions, and GitHub activity signals are all surfaced upfront. Agents have genuine decision-making material without any human intermediary.
Network effects (P5-D: 5/5) The catalog model creates genuine compounding value — more servers make MCPMarket more valuable to agents, which attracts more server publishers. This is a structural moat.
Scoped permissions via MCP protocol (P3-B: 3/5) The underlying MCP protocol requires explicit permission declaration. MCPMarket surfaces this as a discovery signal, giving agents a built-in trust filter.

Critical Gaps

No schema.org markup (P1-A: 2/5) The platform that indexes AI tools for agents is itself invisible to agent-driven AI retrieval. No Offer, SoftwareApplication, or WebAPI schema found anywhere on the site.
No /llms.txt (P1-C: 1/5) MCPMarket — a directory built explicitly for AI agent tool discovery — has no /llms.txt. This is the single biggest irony in any audit conducted. The platform for AI agents cannot be discovered by AI agents via the standard agent discovery protocol.
No scope or limit documentation (P2-B: 1/5) Agents integrating MCP servers via MCPMarket have no platform-level data on rate limits, uptime, or service caps. Each server is a black box from a limits perspective.
No substitution or fallback rules (P2-C: 1/5) If a listed server is deprecated, taken offline, or changes its API, MCPMarket provides no fallback protocol. Agents relying on MCPMarket-sourced tools have no safety net.
No verifiable performance data (P3-A: 1/5) No uptime SLA, no performance benchmarks, no third-party verification for either the platform or the servers it lists.

Priority Actions

Publish /llms.txt immediately — +5 pts · P1 · Effort: Low
Add schema.org/SoftwareApplication markup to all listings — +4 pts · P1 · Effort: Low
Add an account system with saved servers and installation history — +4 pts · P5 · Effort: Med

All 20 Criteria

P1-A Structured Data — 2/5
No schema.org Offer, Product, or SoftwareApplication markup found on directory listing pages. Basic website metadata present but nothing structured for machine consumption.

P1-B Machine-Readable Pricing — 3/5
MCPMarket is a free directory; pricing is clearly "free" but this is communicated in prose only, not tagged or structured. Individual MCP server pricing (where applicable) is listed in HTML text.

P1-C llms.txt / Agent Layer — 1/5
No /llms.txt found. No agent identity document or agent-facing structured overview of the platform exists.

P1-D API / MCP Availability — 4/5
Platform exists specifically to catalog and distribute MCP servers; each listing provides installation commands and configuration snippets. No MCPMarket API for programmatic directory queries found, which limits score.

P1-E Discoverability (GEO) — 3/5
Site content is clear and well-structured for web indexing; MCP server descriptions include technical detail. Not yet optimized for AI retrieval specifically.

P2-A Offer Completeness — 3/5
Directory listings include server name, description, and install instructions but offer details (capabilities, limits, pricing) vary significantly by listing. Not machine-parseable from a single source.

P2-B Scope & Limits — 1/5
No rate limits, usage caps, or service boundaries documented at the platform level. Individual server limits depend entirely on the upstream provider; MCPMarket provides no aggregated limit data.

P2-C Substitution Rules — 1/5
No guidance on what happens when a listed MCP server is unavailable, deprecated, or removed. No fallback or substitution framework.

P2-D Conditional Logic — 2/5
Platform terms are minimal; conditions for listing availability, server uptime, and removal are not disclosed. Agents cannot anticipate condition changes.

P2-E Semantic Precision — 3/5
Server descriptions vary in quality by contributor; some are precise and technical, others are vague. No platform-enforced precision standard.

P3-A Verifiable Performance — 1/5
No uptime data, SLA, or third-party performance verification at the platform level. MCPMarket is a directory, not a host — but no verification of listed servers is provided either.

P3-B Scoped Permissions — 3/5
MCP protocol inherently supports scoped permissions per server; MCPMarket listings document required permissions for each server. Platform-level permission scoping not present.

P3-C Audit Trail — 2/5
No machine-accessible audit log or transaction history. GitHub links for open-source servers provide some version history, but this is not systematic.

P3-D Behavioral Consistency — 4/5
Platform appears stable; listings include version numbers and GitHub commit history. Open-source nature of most entries provides some behavioral consistency signal.

P4-A Friction-Free Activation — 5/5
Zero friction: no signup, no payment, no approval required. Agents (or developers) can discover, review, and install any MCP server directly from the listing. Pure discovery with no gate.

P4-B Agent Decision Signals — 5/5
Install commands are copy-paste ready; server capabilities and required permissions are listed upfront; GitHub stars and activity visible as quality signals. Agents have everything needed to evaluate without human input.

P5-A Integration Depth — 4/5
Once an MCP server is integrated into an agent's configuration, switching requires deliberate re-configuration. MCPMarket is the discovery layer; the installed server creates the actual switching cost. Network effects are strong.

P5-B Agent Memory Layer — 2/5
No account system, no saved favorites, no installation history tracked at the platform level. Agents must re-discover servers from scratch in each session.

P5-C Programmatic Renewal — 2/5
No renewal mechanism exists; MCPMarket is a directory, not a subscription. Updates to installed servers require manual re-discovery or direct GitHub monitoring.

P5-D Compounding Value — 5/5
[Capped at 5] Strong: more MCP servers listed → more agent use cases served → more developers publishing to MCPMarket → more agents adopting it as the canonical discovery layer. Network effects are real and accelerating.

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