Agentic Skills Marketplaces Compared: What Exists, What's Missing, and What Comes Next

The landscape in mid-2026

The agentic skills marketplace space went from zero to fragmented in approximately 18 months. Between late 2024 and mid-2026, at least eight notable platforms launched. None has consolidated the category. The format war (SKILL.md vs. proprietary formats) ended; the distribution war has not.

This comparison is neutral. Skills Warehouse is building an independent marketplace and has an obvious interest in this analysis. Where we have a view, we state it as a view, not as fact.

The platforms

Agensi

Type: Curated marketplace Focus: Agentic skills (SKILL.md-class) Model: 70/30 revenue split (creator keeps 70%) Security: 8-point security scan Delivery: One-command install + early MCP support Positioning: Developer-tools brand

What Agensi does well:

• The strongest curation + security combination in the current market
• Clean developer-focused UX
• The 8-point scan is the current best-practice standard for trust

Limitations:

• Developer brand only — not accessible to non-technical buyers
• Single-platform preference in recommendations
• No physical skills dimension
• No cross-format ingestion (Cursor rules, GPT configs, prompt libraries stay siloed)
• No enterprise procurement tier with indemnification

Assessment: The current best-in-class for developer-focused agentic skills. The benchmark to match on trust, exceed on range and neutrality.

---

Smithery

Type: MCP directory Focus: MCP servers Model: Free listing, no payments Security: None documented Delivery: Directory discovery, not commerce

What Smithery does well:

• Large MCP server catalogue
• Clean discovery UX
• Strong developer community reputation

Limitations:

• No security scanning
• No payments or creator monetisation
• No SKILL.md coverage
• No physical skills
• Discovery tool, not a marketplace

Assessment: One aisle of a warehouse. Useful for MCP discovery; no trust layer; not a commercial platform.

---

PulseMCP

Type: MCP directory and analytics Focus: MCP servers and ecosystem data Model: Free, analytics-first Security: None documented Delivery: Directory

What PulseMCP does well:

• Ecosystem analytics (server growth data, download stats)
• Discovery for MCP servers

Limitations:

• Same gap as Smithery: no security, no payments, no SKILL.md, no physical
• Data product, not a marketplace

Assessment: Valuable for ecosystem intelligence. Not a commercial marketplace.

---

mcpmarket.com

Type: Volume directory Focus: MCP servers Model: Free listing Security: Not documented Scale: 31,000+ skills listed (via scraping)

What mcpmarket does:

• Volume — the largest raw catalogue count in the ecosystem
• GitHub scraper-based ingestion at scale

Limitations:

• Volume without trust is noise
• No security verification
• No provenance
• No payments
• No curation

Assessment: A cataloguing exercise, not a marketplace. The 31,000 number is real; the trust value is near zero.

---

skills.sh

Type: Developer directory Focus: SKILL.md files Model: Listing-based Security: Not documented

Assessment: Early-stage directory. Not yet a commercial product.

---

SkillsMP

Type: Skills marketplace Focus: AI/agentic skills Status: Early stage

Assessment: Insufficient public information to provide a fair comparison.

---

Platform-owned marketplaces (Salesforce AgentExchange, Microsoft Copilot Studio, Google Agentspace)

Type: Enterprise marketplaces Focus: Platform-native skills Model: Revenue share, enterprise procurement

What they do well:

• Enterprise procurement rails
• Native distribution to platform users
• Compliance and security tooling for their own platforms

Fundamental limitation: Platform owners are structurally conflicted. AgentExchange prioritises Salesforce-native skills. Microsoft's marketplace prioritises Copilot-native integrations. These are not neutral catalogues — they are proprietary ecosystems wearing a marketplace label.

The gap they create: Independent, cross-platform skills that work across Claude AND GPT AND Gemini are systematically deprioritised or excluded from platform-owned marketplaces. This is not malicious — it is structural. Platform marketplaces are distribution tools for their own ecosystems, not neutral warehouses.

Assessment: The right channel for platform-native enterprise deployment. The wrong channel for cross-platform skills. The independent seat exists precisely because these platforms cannot fill it.

---

Online learning platforms (Coursera, Udemy, Skillshare, YouTube)

Type: Human skill education Focus: Human learners Model: Subscription or per-course, creator-funded via revenue share

Relevance to this comparison: These platforms own the human skill education market ($222B+ in 2026). They do not compete with agentic skills marketplaces — yet. Their content is potential input for physical skill ingestion (licensed video → structured Skill Card) rather than direct competition.

The January 2026 Coursera-Udemy merger signals consolidation in the human skill education space. The platform is mature; the margins are compressing.

Assessment: Adjacent, not competitive. Potential content supply chain partner, not a rival.

---

The gap matrix

| Need | Agensi | Smithery | mcpmarket | Platform-owned | Available anywhere | |------|--------|----------|-----------|----------------|-------------------| | Curated agentic skills (SKILL.md) | ✅ | ❌ | ❌ | Partial | ✅ | | MCP server discovery | Partial | ✅ | ✅ | ✅ | ✅ | | Security verification | ✅ | ❌ | ❌ | Partial | Partial | | Cross-platform neutrality | Partial | ✅ | ✅ | ❌ | Partial | | Creator payments | ✅ | ❌ | ❌ | ✅ | Partial | | Physical / human skills | ❌ | ❌ | ❌ | ❌ | ❌ | | Enterprise procurement tier | ❌ | ❌ | ❌ | ✅ | ❌ | | Cross-format ingestion | ❌ | ❌ | ❌ | ❌ | ❌ | | Knowledge base / reference | Partial | ❌ | ❌ | ❌ | ❌ | | Non-developer UX | ❌ | ❌ | ❌ | Partial | ❌ |

The gap nobody fills: Neutral + trusted + cross-format + human skills + non-developer accessible. These four requirements together describe a different kind of platform — not a developer tool, not a platform marketplace, but a true warehouse.

What comes next: predictions for 12–24 months

Consolidation. Eight to twelve platforms will not survive. The category will likely consolidate to two or three: one platform-owned (Anthropic or OpenAI first-party), one enterprise (Microsoft or Salesforce), and one independent cross-platform warehouse. The independent seat is the one being contested.

Trust as table stakes. Security incidents from unverified skills will force the ecosystem to demand scan reports and provenance chains. Directories without security verification will either add it or lose enterprise and security-conscious developer trust.

MCP-native delivery as default. The download-install model gives way to live MCP-based access. Marketplaces that don't run their own MCP catalogue infrastructure will become discovery layers feeding someone else's delivery system.

Agent-initiated commerce. As agent payment protocols (x402, AP2) mature, agents will discover, evaluate, and purchase skills autonomously. The marketplace that is machine-readable — structured metadata, security attestations, licensing terms an agent can parse — captures demand that human-facing stores cannot see.

Physical skills as a delayed frontier. None of the current marketplaces has addressed physical skills at scale. The standardisation of procedural human knowledge is a large open problem — and the content layer for embodied AI. Whoever builds the Skill Card library now is planting a flag in the embodied AI era.

---

Skills Warehouse is building the neutral, trust-maximal, cross-format warehouse that this comparison says doesn't yet exist. Publish your skill free →