State of the AI agent ecosystem
What 73,820 indexed agent products look like: types, launches per month, publishers, licences, hosting and data completeness, generated from the AgentSidekick database.
Executive summary
The catalogue holds 73,820 active products; 69,078 of them (93.6%) are MCP servers from the official MCP Registry.
69,228 products have a launch date in the last 12 months. In the three most recent full months 34,611 launched, against 11,912 in the three months before.
45,124 publishers stand behind the catalogue, but the 10 largest account for 15.2% of all products; 40,829 publishers have exactly one.
A licence is known for only 31,724 products (43.0%); among those, 95.8% use MIT, Apache-2.0 or another permissive licence.
41,716 products declare a remote endpoint only, 23,056 run locally only and 4,833 offer both.
66,478 products (90.1%) meet the data-completeness bar; 9,469 of them currently have profile pages.
Every figure above is computed from the database DER
Methodology
How this report was produced, and its limits
Generated on 2026-11-21 from the AgentSidekick database. The latest completed ingestion step finished at 2026-11-21 13:23 UTC. Tables queried: products, companies, sources, product_repos, repositories, ingestion_runs. The catalogue holds 73,820 active products, of which 9,469 meet the data-completeness bar and have a profile page.
Classification method: product type comes from the source (registry entries are MCP servers by definition) or is inferred from repository topics; categories, capabilities, integrations and industries come from keyword rules in the editable taxonomy, so they are derived, not checked by a person. Curated records were researched against the official sources listed on each record.
Known limits:
- Launch dates are the first publish date in the source registry (official MCP Registry) or the GitHub repository creation date. They are not product announcements, and a re-published entry can look like a new launch.
- Only 1 daily metrics snapshot(s) exist in product_metrics, so growth is measured from launch counts, not from changes in stars or downloads.
- Publisher identity is the namespace or GitHub owner. Organisation vs individual comes from the GitHub account type where known; most registry namespaces have no account type and are counted as "unknown".
- Licence comes from the product record or, failing that, the licence GitHub detected in its repository. Registry entries rarely carry one, so the licence mix describes a minority of the catalogue.
- Categories, capabilities and integrations are assigned by keyword rules over names and descriptions (derived). Products that do not name a capability are missed, and generic words can produce false matches.
Note. Claim labels: source = stated by the publisher or registry; curated = checked by us against cited sources; derived = computed by a documented rule; heuristic = a rule of thumb with no measurement behind it. Full method: methodology.
What the catalogue is made of
MCP servers dominate by volume: 69,078 of 73,820 active products, because the official MCP Registry publishes a machine-readable list that can be ingested completely. Every other type is found through GitHub topic search or curated research, so the next largest type, Agent skills, has 2,056 entries. The table is therefore a picture of what is easy to index, not a market-share estimate: an agent skill shared in a single repository folder and a commercial agent platform each count as one product.
Active products by typeDER
Type is a source fact for registry entries and inferred from repository topics for GitHub discoveries.
| Type | Products | Share | With profile page |
|---|---|---|---|
| MCP servers | 69,078 | 93.6% | 5,452 |
| Agent skills | 2,056 | 2.8% | 1,898 |
| Plugins | 1,088 | 1.5% | 722 |
| Infrastructure | 384 | 0.5% | 364 |
| Frameworks | 273 | 0.4% | 248 |
| Rules packs | 267 | 0.4% | 101 |
| Agent projects | 244 | 0.3% | 195 |
| IDE extensions | 148 | 0.2% | 125 |
| Libraries | 116 | 0.2% | 105 |
| Platforms | 47 | 0.1% | 47 |
| AI APIs | 29 | 0.0% | 29 |
| SaaS products | 26 | 0.0% | 26 |
| Developer tools | 23 | 0.0% | 23 |
| Agent SDKs | 15 | 0.0% | 15 |
| Protocols & standards | 14 | 0.0% | 14 |
| Automation platforms | 12 | 0.0% | 12 |
Products by primary sourceSRC
| Source | Products | Share |
|---|---|---|
| Official MCP Registry | 68,059 | 92.2% |
| GitHub topic search | 3,291 | 4.5% |
| SKILL.md files in public GitHub repos | 1,321 | 1.8% |
| Claude plugin marketplace manifests | 742 | 1.0% |
| Curated research | 259 | 0.4% |
| Open VSX Registry | 148 | 0.2% |
What this does not show: products that are listed only in vendor marketplaces (Cursor, VS Code, the GPT Store) are not ingested yet, so IDE extensions and hosted assistants are under-represented next to registry-native MCP servers.
Launches per month
71,424 products (96.8%) carry a launch date. Launches are not evenly spread: of the 69,228 dated in the last 12 months, 67,617 are MCP servers. The three most recent full months (2026-08 to 2026-10) saw 34,611 launches, compared with 11,912 in 2026-05 to 2026-07. The current month (2026-11) is partial and is shown for completeness only.
MCP servers launched per month (registry publish date)SRC
Unit: products per month, 2025-12 to 2026-11.
Show the data as a table
| MCP servers (products) | |
|---|---|
| 2025-12 | 291 |
| 2026-01 | 407 |
| 2026-02 | 1,063 |
| 2026-03 | 1,901 |
| 2026-04 | 2,515 |
| 2026-05 | 2,858 |
| 2026-06 | 3,734 |
| 2026-07 | 4,963 |
| 2026-08 | 6,434 |
| 2026-09 | 11,718 |
| 2026-10 | 16,071 |
| 2026-11 | 15,662 |
Other products launched per month (repository creation date)SRC
Unit: products per month, 2025-12 to 2026-11.
- Agent skills
- Plugins
- Everything else
Show the data as a table
| Agent skills (products) | Plugins (products) | Everything else (products) | |
|---|---|---|---|
| 2025-12 | 25 | 12 | 22 |
| 2026-01 | 72 | 29 | 39 |
| 2026-02 | 63 | 27 | 88 |
| 2026-03 | 85 | 42 | 118 |
| 2026-04 | 95 | 36 | 76 |
| 2026-05 | 41 | 17 | 68 |
| 2026-06 | 49 | 19 | 59 |
| 2026-07 | 39 | 17 | 48 |
| 2026-08 | 41 | 11 | 74 |
| 2026-09 | 40 | 24 | 39 |
| 2026-10 | 58 | 61 | 40 |
| 2026-11 | 13 | 9 | 15 |
Read the MCP line with care. A registry publish date is when the entry first appeared in the registry, which for many servers is months after the code existed. Repository creation dates for skills and plugins have the opposite bias: they predate the moment the repository became a skill collection. Neither line measures adoption.
Who publishes
45,124 distinct publishers have at least one active product. Concentration is high at the top: the 10 largest publishers account for 11,203 products (15.2%) and the top 1% of publishers (451 accounts) for 30.6%. At the other end, 40,829 publishers (90.5%) have a single product. Several of the largest namespaces publish hundreds or thousands of near-identical registry entries, so a count of entries overstates how many distinct tools exist.
Largest publishers by product countDER
| Publisher | Account type | Products | Share of catalogue | With profile page |
|---|---|---|---|---|
| mrfentmen | individual | 2,740 | 3.7% | 13 |
| sadri-dridi | individual | 2,366 | 3.2% | 13 |
| pipeworx-io | organization | 1,914 | 2.6% | 46 |
| mcp-dir | organization | 1,120 | 1.5% | 9 |
| usefulapi.io | organization | 1,095 | 1.5% | 23 |
| CloudCorpRecords | individual | 562 | 0.8% | 25 |
| theluckystrike | individual | 424 | 0.6% | 2 |
| cotlet.dev | organization | 398 | 0.5% | 17 |
| Evozim | individual | 375 | 0.5% | 0 |
| smithery.ai | organization | 209 | 0.3% | 55 |
| CSOAI-ORG | individual | 177 | 0.2% | 3 |
| cyanheads | individual | 162 | 0.2% | 16 |
| jmshinhwa | individual | 155 | 0.2% | 3 |
| ccplugins | organization | 150 | 0.2% | 143 |
| mambabuilt.com | organization | 148 | 0.2% | 10 |
Products by publisher account typeDER
| Account type | Publishers | Products | Share of products |
|---|---|---|---|
| organization | 28,889 | 42,256 | 57.2% |
| individual | 11,625 | 25,222 | 34.2% |
| unknown | 4,610 | 6,342 | 8.6% |
Organisation accounts publish 42,256 products and individual accounts 25,222, but 6,342 products (8.6%) come from namespaces whose account type is unknown. That gap matters for buyers: "published by an organisation" is a weak trust signal on its own, and for most of the catalogue it cannot be stated at all.
Licences and hosting
Licence information is the thinnest field in the catalogue. Only 31,724 products have a known licence; 42,096 (57.0%) have none on record, usually because the registry entry does not state one and its repository has not been fetched or has no licence file. Among known licences, permissive ones (MIT, Apache-2.0, BSD and similar) make up 95.8%.
Most common licences (products with a known licence)SRC
Unit: products. Top 12 shown.
Show the data as a table
| products | |
|---|---|
| MIT | 25,591 |
| Apache-2.0 | 4,575 |
| AGPL-3.0 | 781 |
| GPL-3.0 | 236 |
| BSD-3-Clause | 76 |
| CC-BY-4.0 | 65 |
| ISC | 63 |
| CC0-1.0 | 56 |
| MIT-0 | 51 |
| MPL-2.0 | 46 |
| GPL-2.0 | 40 |
| WTFPL | 15 |
Hosting modelDER
Unit: products. Top 4 shown.
Show the data as a table
| products | |
|---|---|
| remote | 41,716 |
| local | 23,056 |
| both | 4,833 |
| unknown | 4,215 |
Hosting is derived from what a registry entry declares: a remote endpoint means "remote", a package means "local", both means "both". Remote-only entries (41,716) outnumber local-only ones (23,056), which is the reverse of the early MCP ecosystem where nearly every server was a local process. 4,215 products have no hosting declaration, mostly libraries, frameworks and skills where the question does not apply.
Data completeness, and why it matters
Every product gets a completeness score from 0 to 1 based on which fields are filled (description, repository, licence, install information and so on). 66,478 products reach 0.6 or are curated; only those are eligible for a profile page, and the build caps pages at 9,469. The remaining 7,342 products are counted in every total on this site but do not get their own page, because a page with a name and nothing else helps nobody.
Products by data-completeness scoreDER
Unit: products. Top 5 shown.
Show the data as a table
| products | |
|---|---|
| 0–0.2 | 32 |
| 0.8–1.0 | 30,176 |
| 0.6–0.8 | 36,301 |
| 0.4–0.6 | 6,349 |
| 0.2–0.4 | 962 |
The practical consequence: lists and comparisons on this site are drawn from the 9,469 profiled products, while market counts (segments, launches, publishers) use all 73,820. The most common type, MCP servers, has 5,452 profiled out of 69,078 (7.9%), so for that type a "top 10" list is a selection from a much larger, mostly undocumented pool.
Recommendations
Editorial: the judgement of AgentSidekick, drawn from the figures above. Not a measurement, not investment advice.
- Editorial
treat raw MCP server counts as an upper bound. The 10 largest publishers alone account for 15.2% of the catalogue, so deduplicate by publisher before quoting "number of tools".
- Editorial
if you publish an agent product, state a licence. 57.0% of the catalogue has none on record, so a clear licence is an easy way to stand out in evaluations.
- Editorial
publishers whose entries fall below the completeness bar (7,342 products) can get a profile by adding a description, repository link and install information to their registry entry.
- Editorial
compare launch momentum only between like sources. The 34,611 vs 11,912 launch comparison mixes registry publish dates with repository creation dates.
Sources
- Official MCP Registry: registry.modelcontextprotocol.io (external)checked
- GitHub topic search: docs.github.com (external)checked
- GitHub repository metadata: docs.github.com (external)checked
- GitHub REST API: repositories: docs.github.com (external)checked
- Open VSX Registry: open-vsx.org (external)checked
- AgentSidekick methodology: agentsidekick.dev (external)checked
Data behind this report
- Index of profiled products: /api/dataset/products.jsonJSON, regenerated each build
- Per-segment counts and launch momentum: /api/dataset/segments.jsonJSON, regenerated each build
Cite this report
State of the AI agent ecosystem. AgentSidekick, 2026-11-21. https://www.agentsidekick.dev/reports/state-of-the-agent-ecosystem/
Figures change when the database is refreshed; cite the date shown.