Blog
hedwigaihedwigaihedwigai

Agent tool usage · week to 21 October 2026

Which tools do AI agents use, by marketplace calls, SDK downloads and MCP installs?

No single source answers this question. We compared four public sources for the week to 21 October 2026: calls on the Monid marketplace, MCP server uses on Smithery, and SDK downloads on npm and PyPI. Every table is in the workbook behind this post.

What the sources show, in six lines:

  1. Web search and social media take 76% of Monid calls. One provider takes most of each: TinyFish in search and TikHub in social.
  2. 75% of Monid calls are free. Agents made 5.78M calls through Monid in the week. Only 1.42M of them were paid.
  3. A few accounts make most of the calls. Endpoints that average 100 or more calls per account each week carry 84% of the volume.
  4. The largest weekly swings come from small providers. Providers with under 5K calls changed by a median of +101% in a week. The largest providers changed by +28%.
  5. The sources rank providers differently. On Monid, TinyFish has 25 times the calls of Firecrawl. On PyPI, Firecrawl has more than 100 times the downloads of TinyFish.
  6. Tavily and Brave Search do not appear on Monid. Tavily has 1.71M PyPI downloads a week. Brave Search has 119K uses on Smithery, the most of any server we found.

The data

Four public sources and what each counts

Each source counts something different. Monid counts calls. Smithery counts uses of MCP servers. npm and PyPI count downloads of code. None of them counts everything.

Monid

A marketplace that sells tool calls to AI agents

Counts
Every call an agent makes through Monid, paid or free, and the accounts that called
Cannot see
Calls made straight to a provider, without Monid
Used here
45 providers, 327 endpoints, 5.78M calls in the week
Leader
TinyFish, 33% of ranked calls

Smithery

The largest registry of MCP servers

Counts
How often each server it lists or hosts has been used, by its own count
Cannot see
Servers that are not on Smithery
Used here
1,498 servers found by 36 searches, with 1.04M uses
Leader
Brave Search, 11% of the uses found

npm

The JavaScript package registry

Counts
Downloads of each package in the last week
Cannot see
Whether the code ever calls the tool. Automated builds count as downloads.
Used here
The official SDKs of nine providers
Leader
Firecrawl, 1.13M a week

PyPI

The Python package registry

Counts
Downloads of each package in the last week, mirrors included
Cannot see
Whether the code ever calls the tool. Automated builds count as downloads.
Used here
The official SDKs of eight providers
Leader
ElevenLabs, 2.29M a week, then Tavily, 1.71M

1 · Ranking

Top 12 providers by Monid calls

These twelve providers have the most calls on Monid in the week. The bar shows calls. The next columns show how many accounts called each provider and how its calls changed from the week before.

Monid ranked calls, week to 21 October 2026 (a free call counts as one fifth)
#ProviderCallsWeek
1TinyFish
763K
30%
2TikHub
531K
6%
3AnyAPI
251K
new
4Litescrape
173K
5%
5Context.dev
148K
40%
6Hunter
78.9K
28%
7Apollo
61.8K
64%
8DataForSEO
60.0K
28%
9MrScraper
47.3K
101%
10Ploid
41.3K
26%
11Firecrawl
30.6K
47%
12GPT Images
14.8K
82%

TinyFish has 44% more calls than TikHub, the second provider. It also has the widest reach, with 368 accounts. Among these twelve, Apollo, Hunter and Context.dev come next by reach, with 280 to 329 accounts each. They rank lower by calls. All 45 providers are in the workbook.

2 · Categories

Monid calls by category and provider

Monid files every endpoint under a category. Adding up the calls shows what agents use tools for. Each bar is one category. Its length is the category’s share of all calls. Its segments are the providers that take those calls.

Monid ranked calls by category and provider, week to 21 October 2026
Web search927K · 40.4%

TinyFish 82% · Litescrape 6% · MrScraper 4% · others 7%

Social813K · 35.5%

TikHub 64% · AnyAPI 29% · Litescrape 5% · others 2%

Lead gen207K · 9.0%

Hunter 38% · Apollo 30% · Ploid 20% · others 12%

Web scraping171K · 7.5%

Context.dev 79% · Firecrawl 17% · DataForSEO 2% · others 2%

E-commerce & apps83.1K · 3.6%

Litescrape 84% · TikHub 8% · DataForSEO 4% · others 4%

Gen AI37.5K · 1.6%

GPT Images 39% · ElevenLabs 18% · TypeSafe 16% · others 27%

SEO & GEO24.8K · 1.1%

DataForSEO 65% · cloro 15% · Litescrape 10% · others 10%

Travel, property & jobs17.8K · 0.8%

DataForSEO 38% · MrScraper 35% · Zillow 18% · others 9%

Other11.6K · 0.5%

Simple FS 49% · AnyAPI 26% · Saperly 10% · others 15%

Most categories have one leader. TinyFish takes 82% of web search. Context.dev takes 79% of web scraping. TikHub takes 64% of social, and AnyAPI takes most of the rest. Lead gen is the exception. Hunter, Apollo and Ploid share it more evenly.

3 · Paid and free

Paid and free calls on Monid

7 days5.78M calls
paid 24.6%free 75.4%
30 days21.0M calls
paid 21.2%free 78.8%

Monid ranks providers by paid calls plus one fifth of free calls. In the week that total was 2.29M. This weighting is a choice about what to rank. It is not a measurement. Under it, a provider’s place depends on its mix of free and paid calls as much as on its volume.

All calls grew 22.3% in a week, from 4.73M to 5.78M.

4 · Accounts

Calls per account

Share of Monid calls, by how many calls each account made to an endpoint in the week
1,000 or more
30.9%
100 to 999
53.5%
10 to 99
15.2%
under 10
0.5%

A thousand calls a week from one account is a pipeline, not a person. It may be a scheduled job, a crawler or a batch. Such use is real. It is also concentrated, and it can appear or vanish with one customer.

Monid counts accounts for each endpoint. It does not say whether an account that calls two endpoints is counted once or twice. A provider’s reach here is therefore at least the count of its largest endpoint.

5 · Patterns

Reach and intensity across all 327 endpoints

Each dot is one Monid endpoint. Further right means more accounts call it. Higher means each account calls it more often. Both scales are logarithmic, so each step is ten times the last.

Monid endpoints, week to 21 October 2026: accounts against calls per account
Social Search, scraping and lead gen Every other category
1101001101001,000accounts calling the endpointE-commerce & apps: 20 accounts, 1,900 calls eachE-commerce & apps: 28 accounts, 1,068 calls eachGen AI: 39 accounts, 175 calls eachGen AI: 25 accounts, 235 calls eachGen AI: 31 accounts, 174 calls eachE-commerce & apps: 11 accounts, 365 calls eachGen AI: 110 accounts, 32 calls eachTravel, property & jobs: 12 accounts, 269 calls eachTravel, property & jobs: 13 accounts, 246 calls eachTravel, property & jobs: 24 accounts, 125 calls eachOther: 313 accounts, 9 calls eachSEO & GEO: 55 accounts, 48 calls eachOther: 18 accounts, 137 calls eachTravel, property & jobs: 4 accounts, 615 calls eachGen AI: 80 accounts, 30 calls eachSEO & GEO: 32 accounts, 68 calls eachOther: 317 accounts, 7 calls eachSEO & GEO: 19 accounts, 105 calls eachE-commerce & apps: 94 accounts, 21 calls eachSEO & GEO: 45 accounts, 43 calls eachE-commerce & apps: 72 accounts, 24 calls eachTravel, property & jobs: 6 accounts, 265 calls eachE-commerce & apps: 4 accounts, 392 calls eachGen AI: 46 accounts, 34 calls eachTravel, property & jobs: 4 accounts, 374 calls eachGen AI: 21 accounts, 67 calls eachSEO & GEO: 208 accounts, 7 calls eachSEO & GEO: 19 accounts, 72 calls eachGen AI: 32 accounts, 40 calls eachSEO & GEO: 24 accounts, 54 calls eachGen AI: 12 accounts, 107 calls eachSEO & GEO: 127 accounts, 10 calls eachSEO & GEO: 14 accounts, 87 calls eachSEO & GEO: 30 accounts, 40 calls eachGen AI: 11 accounts, 109 calls eachSEO & GEO: 36 accounts, 33 calls eachE-commerce & apps: 18 accounts, 62 calls eachGen AI: 35 accounts, 29 calls eachSEO & GEO: 56 accounts, 18 calls eachSEO & GEO: 16 accounts, 62 calls eachGen AI: 9 accounts, 107 calls eachTravel, property & jobs: 9 accounts, 104 calls eachSEO & GEO: 15 accounts, 60 calls eachGen AI: 129 accounts, 7 calls eachSEO & GEO: 46 accounts, 18 calls eachSEO & GEO: 149 accounts, 6 calls eachE-commerce & apps: 15 accounts, 54 calls eachFinance data: 5 accounts, 153 calls eachGen AI: 4 accounts, 186 calls eachGen AI: 34 accounts, 22 calls eachGen AI: 47 accounts, 16 calls eachOther: 49 accounts, 15 calls eachTravel, property & jobs: 4 accounts, 178 calls eachOther: 5 accounts, 135 calls eachE-commerce & apps: 20 accounts, 33 calls eachE-commerce & apps: 7 accounts, 90 calls eachSEO & GEO: 4 accounts, 157 calls eachTravel, property & jobs: 26 accounts, 24 calls eachSEO & GEO: 2 accounts, 302 calls eachE-commerce & apps: 6 accounts, 98 calls eachTravel, property & jobs: 104 accounts, 5 calls eachE-commerce & apps: 10 accounts, 56 calls eachSEO & GEO: 14 accounts, 40 calls eachE-commerce & apps: 30 accounts, 18 calls eachOther: 10 accounts, 53 calls eachSEO & GEO: 7 accounts, 73 calls eachGen AI: 5 accounts, 97 calls eachOther: 4 accounts, 113 calls eachGen AI: 49 accounts, 9 calls eachE-commerce & apps: 34 accounts, 13 calls eachGen AI: 26 accounts, 15 calls eachOther: 4 accounts, 90 calls eachE-commerce & apps: 5 accounts, 70 calls eachOther: 3 accounts, 110 calls eachE-commerce & apps: 26 accounts, 12 calls eachGen AI: 13 accounts, 15 calls eachSEO & GEO: 12 accounts, 16 calls eachOther: 16 accounts, 11 calls eachGen AI: 24 accounts, 7 calls eachOther: 2 accounts, 60 calls eachSEO & GEO: 28 accounts, 4 calls eachOther: 2 accounts, 2 calls eachWeb search: 368 accounts, 1,522 calls eachWeb search: 324 accounts, 628 calls eachWeb scraping: 280 accounts, 407 calls eachSocial: 62 accounts, 961 calls eachWeb search: 125 accounts, 448 calls eachSocial: 135 accounts, 341 calls eachSocial: 51 accounts, 871 calls eachSocial: 38 accounts, 1,102 calls eachLead gen: 329 accounts, 121 calls eachWeb search: 92 accounts, 432 calls eachSocial: 44 accounts, 823 calls eachLead gen: 139 accounts, 254 calls eachSocial: 42 accounts, 794 calls eachSocial: 69 accounts, 377 calls eachSocial: 25 accounts, 1,031 calls eachSocial: 114 accounts, 226 calls eachSocial: 53 accounts, 474 calls eachWeb scraping: 170 accounts, 127 calls eachSocial: 22 accounts, 932 calls eachSocial: 236 accounts, 82 calls eachWeb scraping: 78 accounts, 241 calls eachSocial: 69 accounts, 254 calls eachSocial: 34 accounts, 426 calls eachSocial: 32 accounts, 445 calls eachLead gen: 195 accounts, 73 calls eachLead gen: 147 accounts, 94 calls eachSocial: 93 accounts, 145 calls eachSocial: 27 accounts, 471 calls eachSocial: 36 accounts, 327 calls eachWeb search: 148 accounts, 76 calls eachWeb search: 160 accounts, 66 calls eachWeb search: 88 accounts, 120 calls eachLead gen: 75 accounts, 140 calls eachLead gen: 321 accounts, 32 calls eachSocial: 42 accounts, 222 calls eachWeb search: 28 accounts, 323 calls eachSocial: 168 accounts, 54 calls eachWeb search: 265 accounts, 34 calls eachSocial: 25 accounts, 343 calls eachLead gen: 45 accounts, 190 calls eachSocial: 34 accounts, 240 calls eachSocial: 61 accounts, 132 calls eachWeb scraping: 18 accounts, 441 calls eachLead gen: 35 accounts, 225 calls eachSocial: 41 accounts, 189 calls eachSocial: 241 accounts, 32 calls eachSocial: 96 accounts, 78 calls eachSocial: 170 accounts, 44 calls eachSocial: 33 accounts, 221 calls eachLead gen: 64 accounts, 113 calls eachSocial: 6 accounts, 1,198 calls eachLead gen: 323 accounts, 22 calls eachSocial: 19 accounts, 356 calls eachSocial: 76 accounts, 89 calls eachLead gen: 99 accounts, 67 calls eachSocial: 7 accounts, 940 calls eachSocial: 10 accounts, 648 calls eachSocial: 6 accounts, 967 calls eachSocial: 116 accounts, 48 calls eachSocial: 3 accounts, 1,808 calls eachSocial: 20 accounts, 266 calls eachLead gen: 55 accounts, 96 calls eachSocial: 67 accounts, 77 calls eachSocial: 171 accounts, 30 calls eachSocial: 54 accounts, 94 calls eachSocial: 137 accounts, 37 calls eachSocial: 33 accounts, 152 calls eachSocial: 37 accounts, 131 calls eachSocial: 86 accounts, 56 calls eachSocial: 158 accounts, 30 calls eachLead gen: 46 accounts, 102 calls eachWeb search: 41 accounts, 115 calls eachLead gen: 59 accounts, 74 calls eachSocial: 42 accounts, 101 calls eachSocial: 10 accounts, 404 calls eachLead gen: 38 accounts, 100 calls eachSocial: 6 accounts, 631 calls eachLead gen: 8 accounts, 471 calls eachSocial: 7 accounts, 528 calls eachSocial: 104 accounts, 35 calls eachSocial: 17 accounts, 203 calls eachSocial: 33 accounts, 103 calls eachSocial: 34 accounts, 94 calls eachLead gen: 52 accounts, 60 calls eachSocial: 50 accounts, 60 calls eachLead gen: 25 accounts, 119 calls eachSocial: 33 accounts, 90 calls eachSocial: 52 accounts, 56 calls eachSocial: 27 accounts, 105 calls eachSocial: 22 accounts, 128 calls eachSocial: 16 accounts, 176 calls eachSocial: 66 accounts, 42 calls eachSocial: 103 accounts, 27 calls eachSocial: 7 accounts, 385 calls eachSocial: 7 accounts, 379 calls eachSocial: 17 accounts, 153 calls eachSocial: 24 accounts, 108 calls eachSocial: 14 accounts, 174 calls eachSocial: 10 accounts, 238 calls eachWeb scraping: 48 accounts, 48 calls eachSocial: 65 accounts, 34 calls eachSocial: 47 accounts, 45 calls eachSocial: 39 accounts, 50 calls eachSocial: 64 accounts, 30 calls eachWeb scraping: 36 accounts, 53 calls eachLead gen: 52 accounts, 36 calls eachSocial: 5 accounts, 366 calls eachLead gen: 39 accounts, 47 calls eachSocial: 7 accounts, 256 calls eachSocial: 17 accounts, 105 calls eachWeb search: 67 accounts, 26 calls eachSocial: 7 accounts, 251 calls eachLead gen: 43 accounts, 41 calls eachWeb search: 50 accounts, 35 calls eachWeb search: 41 accounts, 40 calls eachSocial: 68 accounts, 24 calls eachSocial: 91 accounts, 17 calls eachSocial: 8 accounts, 196 calls eachSocial: 68 accounts, 23 calls eachSocial: 19 accounts, 82 calls eachSocial: 66 accounts, 23 calls eachSocial: 70 accounts, 22 calls eachSocial: 14 accounts, 105 calls eachWeb search: 29 accounts, 50 calls eachSocial: 32 accounts, 44 calls eachSocial: 11 accounts, 126 calls eachSocial: 4 accounts, 346 calls eachSocial: 9 accounts, 153 calls eachLead gen: 18 accounts, 74 calls eachSocial: 5 accounts, 264 calls eachSocial: 33 accounts, 39 calls eachSocial: 2 accounts, 643 calls eachSocial: 3 accounts, 422 calls eachLead gen: 16 accounts, 79 calls eachSocial: 13 accounts, 97 calls eachSocial: 17 accounts, 70 calls eachWeb search: 45 accounts, 26 calls eachSocial: 113 accounts, 10 calls eachWeb search: 33 accounts, 35 calls eachSocial: 42 accounts, 27 calls eachSocial: 20 accounts, 55 calls eachSocial: 8 accounts, 138 calls eachSocial: 42 accounts, 26 calls eachLead gen: 18 accounts, 60 calls eachSocial: 25 accounts, 43 calls eachWeb search: 98 accounts, 11 calls eachLead gen: 8 accounts, 130 calls eachSocial: 4 accounts, 259 calls eachSocial: 17 accounts, 59 calls eachWeb scraping: 26 accounts, 38 calls eachSocial: 35 accounts, 28 calls eachSocial: 17 accounts, 58 calls eachSocial: 49 accounts, 20 calls eachSocial: 20 accounts, 49 calls eachSocial: 28 accounts, 34 calls eachSocial: 29 accounts, 32 calls eachSocial: 21 accounts, 44 calls eachSocial: 8 accounts, 113 calls eachLead gen: 48 accounts, 18 calls eachSocial: 26 accounts, 33 calls eachSocial: 54 accounts, 16 calls eachSocial: 5 accounts, 169 calls eachSocial: 34 accounts, 25 calls eachSocial: 23 accounts, 36 calls eachSocial: 13 accounts, 62 calls eachSocial: 58 accounts, 14 calls eachLead gen: 45 accounts, 17 calls eachSocial: 20 accounts, 39 calls eachSocial: 11 accounts, 70 calls eachLead gen: 38 accounts, 20 calls eachLead gen: 21 accounts, 35 calls eachSocial: 34 accounts, 22 calls eachSocial: 44 accounts, 16 calls eachSocial: 3 accounts, 239 calls eachSocial: 31 accounts, 23 calls eachSocial: 49 accounts, 14 calls eachSocial: 5 accounts, 135 calls eachLead gen: 19 accounts, 35 calls eachWeb search: 23 accounts, 29 calls eachWeb search: 15 accounts, 44 calls eachSocial: 11 accounts, 59 calls eachSocial: 48 accounts, 13 calls eachLead gen: 31 accounts, 19 calls eachSocial: 47 accounts, 13 calls eachSocial: 40 accounts, 15 calls eachSocial: 26 accounts, 22 calls eachSocial: 9 accounts, 62 calls eachSocial: 22 accounts, 25 calls eachSocial: 6 accounts, 93 calls eachWeb scraping: 25 accounts, 22 calls eachWeb scraping: 21 accounts, 26 calls eachSocial: 9 accounts, 61 calls eachSocial: 33 accounts, 16 calls eachLead gen: 1 accounts, 542 calls eachSocial: 7 accounts, 77 calls eachWeb scraping: 22 accounts, 24 calls eachSocial: 6 accounts, 87 calls eachSocial: 13 accounts, 40 calls eachLead gen: 26 accounts, 20 calls eachSocial: 10 accounts, 51 calls eachWeb scraping: 23 accounts, 22 calls eachSocial: 6 accounts, 82 calls eachWeb scraping: 33 accounts, 15 calls eachSocial: 26 accounts, 18 calls eachSocial: 18 accounts, 26 calls eachSocial: 50 accounts, 9 calls eachSocial: 25 accounts, 16 calls eachWeb search: 12 accounts, 31 calls eachLead gen: 3 accounts, 116 calls eachSocial: 15 accounts, 22 calls eachSocial: 22 accounts, 15 calls eachWeb search: 14 accounts, 23 calls eachLead gen: 17 accounts, 18 calls eachLead gen: 49 accounts, 6 calls eachSocial: 23 accounts, 12 calls eachWeb search: 12 accounts, 23 calls eachLead gen: 19 accounts, 15 calls eachWeb scraping: 36 accounts, 7 calls eachWeb scraping: 27 accounts, 10 calls eachLead gen: 13 accounts, 19 calls eachSocial: 34 accounts, 7 calls eachWeb scraping: 14 accounts, 16 calls eachLead gen: 18 accounts, 11 calls eachSocial: 26 accounts, 7 calls eachSocial: 3 accounts, 59 calls eachLead gen: 28 accounts, 6 calls eachSocial: 3 accounts, 51 calls eachSocial: 19 accounts, 8 calls eachLead gen: 6 accounts, 15 calls eachSocial: 17 accounts, 5 calls eachLead gen: 9 accounts, 7 calls eachWeb scraping: 18 accounts, 3 calls eachWeb search: 13 accounts, 4 calls eachSocial: 5 accounts, 9 calls eachWeb scraping: 4 accounts, 7 calls eachSocial: 2 accounts, 15 calls eachLead gen: 14 accounts, 1 calls eachWeb search: 10 accounts, 2 calls eachWeb search: 13 accounts, 1 calls eachLead gen: 10 accounts, 1 calls eachSocial: 3 accounts, 3 calls eachLead gen: 9 accounts, 1 calls eachLead gen: 10 accounts, 1 calls eachLead gen: 9 accounts, 0 calls eachWeb scraping: 3 accounts, 0 calls each

Vertical axis: calls per account in the week

Most endpoints are small. 89% of them have fewer than 100 accounts. The eight endpoints that reach 280 accounts or more are general tools: web search and page fetch, scraping, people and email lookup, and file tools. Social data is different. It has 155 endpoints, almost half of all, and none reaches more than 241 accounts. Each social endpoint is called harder: a median of 77 calls per account, against 35 for search, scraping and lead gen.

Median calls per account, by category (categories with five or more endpoints)
Travel, property & jobs
212
Social
77
E-commerce & apps
59
Other
56
SEO & GEO
45
Lead gen
41
Web search
35
Gen AI
34
Web scraping
23

The median social endpoint gets 77 calls from each account in a week. The median lead gen endpoint gets 41. Travel, property and jobs has the highest median, from only 10 endpoints.

6 · Concentration

How concentrated each source is

A source is concentrated when a few items take most of its volume. Each bar shows the share held by the largest item, the largest three and the largest ten.

Share of each source’s volume held by its largest items
Monid45 providers

top 1 33% · top 3 67% · top 10 94%

Monid327 endpoints

top 1 24% · top 3 38% · top 10 53%

Smithery1,498 MCP servers found

top 1 11% · top 3 27% · top 10 46%

Monid is the most concentrated. Its ten largest providers take 94% of all calls. Smithery is far more spread out. Its ten most-used servers take 46% of the uses we found. Most of Smithery is quiet: 1,091 of the 1,498 servers we found have no recorded use.

7 · Change

Weekly change by provider size

Each dot is one Monid provider, placed by how much its calls changed from the week before. Rows group providers by size. Changes beyond +200% are counted at the right edge.

Week-on-week change in Monid calls, by provider size
50K calls or more7 providers · median +28%

0 doubled or more · 0 halved or worse

5K to 50K calls11 providers · median -8%

2 doubled or more · 0 halved or worse

under 5K calls25 providers · median +101%
+9 beyond

13 doubled or more · 5 halved or worse

−100%0+100%+200%

Large providers change slowly. Small ones swing hard in both directions. Among providers with under 5K calls, 13 of 25 at least doubled, and 5 lost half or more. Zillow rose 12,985% to 3,193 calls from 13 accounts. A large percentage on a small base usually means one new customer.

Fastest-rising Monid endpoints
EndpointThis weekLast weekAccountsRise
Litescrape app-store/reviews29.9K5242857×
Litescrape app-store/product38.0K8672044×
Hunter discover7,2551,369645.3×
TikHub Instagram post comments33.3K7,845424.2×
Apollo people/match39.9K13.9K3292.9×

Calls to two Litescrape App Store endpoints rose 44-fold and 57-fold. These calls came from only 20 and 28 accounts respectively. Apollo’s near-tripling came from 329 accounts. That is broad adoption, and it is the more meaningful signal.

8 · Sources compared

The same providers in each source

These eight search and data providers appear in at least two sources. Each column below is scaled to its own largest value, because the sources count different things.

Monid calls, PyPI downloads and Smithery uses, week to 21–22 October 2026
Monid callsPyPI downloadsSmithery uses
TinyFish
763K
10.9K
no official server
DataForSEO
60.0K
55.6K
not found
Firecrawl
30.6K
1.33M
not found
Exa
12.2K
1.57M
9,181
ElevenLabs
6,574
2.29M
not found
Tavilynot listed
1.71M
1,575
Brave Searchnot listedno Python SDK
119K
SerpApinot listed
336K
not found
ProviderMonid calls (rank)npm / weekPyPI / weekSmithery uses
TinyFish763K (#1)1010.9Kno official server
DataForSEO60.0K (#8)22.6K55.6Knot found
Firecrawl30.6K (#11)1.13M1.33Mnot found
Exa12.2K (#14)527K1.57M9,181
ElevenLabs6,574 (#17)550K2.29Mnot found
Tavilynot listed228K1.71M1,575
Brave Searchnot listed12.4Kno Python SDK119K
SerpApinot listed76.1K336Knot found

not listed: Monid does not carry this provider. not found: Smithery’s search for its name returned no server.

The pictures are almost opposite. TinyFish leads Monid by a wide margin and barely appears in downloads. Firecrawl, Exa and Tavily lead downloads and sit low on Monid or are not listed. Neither picture is wrong. Monid shows what agents buy through a marketplace. Downloads show what developers add to their own code.

Downloads have flaws too. They count automated test and build runs. They do not count people. A package name can also belong to someone else. On npm, @apollo/client has 3.2M downloads a week. It is the Apollo GraphQL client and has nothing to do with Apollo.io.

9 · MCP servers

MCP servers by use on Smithery

We searched Smithery for 36 kinds of tool and merged the results by server. Smithery verifies who publishes some servers. The chart shows the twelve most-used verified servers.

Uses of verified MCP servers on Smithery, by Smithery’s own count
Brave Search
119K
Paper Search
103K
Gmail
57.7K
Google Sheets
56.1K
PubMed
43.3K
Google Calendar
17.8K
Slack
12.1K
Exa Search
9,181
Google Drive
8,569
Google BigQuery
6,986
Google Compute Engine
6,365
GitHub
6,059

Search leads here too: Brave Search and Paper Search are first and second, and Exa is eighth. The rest are office tools (Gmail, Google Sheets, Google Calendar, Slack and Google Drive), research data (PubMed) and developer tools (Google BigQuery, Google Compute Engine and GitHub).

This mix differs from Monid’s. On Monid, agents buy search, social media data and sales leads. On Smithery, people connect search, email, documents and research data. Marketplace calls lean towards collecting data in bulk. MCP servers lean towards everyday work.

Some unverified servers report high counts too. Web Scout, for example, reports 19.6K uses. We leave them out of the chart because their counts cannot be checked.

10 · Reading

What each source can and cannot see

The questionSourceWhat it cannot see
Which tools do agents pay for through a marketplace?MonidCalls made straight to a provider, without Monid
Which tools do developers add to their code?npm and PyPIWhether the code ever calls the tool. Automated builds also count as downloads.
Which tools do people connect to their AI assistant?SmitheryServers that are not on Smithery

For any ranking of this kind, read accounts before calls. Read raw figures beside ranked ones. Read every percentage next to the count it started from.

Method

How the data was collected

Everything here came through the shell that your agent gets from hedwigai, on 23 October 2026. Each answer names its source. Each measurement is also kept, so the same commands next week give a trend through series read. The full tables are in the workbook behind this post.

$ agentrank monid providers --limit 50
$ agentrank monid endpoints --limit 400
$ agentrank smithery "web search" --limit 50   # one of 36 searches
$ registry pypi tavily-python --history true
$ registry npm exa-js --history true

$ series read tinyfish --metric calls \
    | jq -c '.[] | [.labels.window, .labels.endpoint, .points[-1].v]'
["30d",null,2789537.0]
["7d",null,763418.0]
["7d","fetch",203332.0]
["7d","search",560086.0]

Limits. This is one week and one snapshot. Monid does not document its account field. We read it as the number of distinct accounts that called each endpoint. Smithery’s search is fuzzy, and Smithery defines its use count itself. Our Smithery list covers the servers that 36 searches returned, not every server. None of these figures is a market share.

Add hedwigai to your MCP client from mcp.hedwigai.com, or sign in with a work email. Turn the programs on with programs enable agentrank and programs enable registry, then ask your agent what a tool is really used for.

More head-to-head evals

How we made a long-running AI agent for knowledge work 4× fasterOne cited knowledge work task through 13 experiments: 937 s to 233 s, with the model held constant.
MeasureFirst runBest since
Landscape, seconds937233
Memo, seconds759602
Read the reportTen models plan a change to a spreadsheet: which do it in under a minute, traced call by callTen models drive one agent harness through the same spreadsheet request; six write the plan, timed and traced call by call.
MeasureQwen3 235BGLM-5.2Gemini 3.8 FlashClaude Opus 5
Seconds to the plan, median27485794
Plans written, of three runs2333
Read the comparisonSubQ, jev, Perplexity's Decider and Bespoke's Nimble on biotech diligence memos: finding the sentences a column is aboutFour decision models judge every sentence of ten biotech diligence memos for five columns, scored against sentences a person marked.
MeasureSubQjevDeciderNimble
Found, mean of five columns at section length64%65%66%68%
Mechanism of action, found at section length57%51%61%54%
Read the comparisonLeads with a reason to write: hedwigai ask, Apollo and treg on one brief40 leads each, every one checked: the right person, an inbox that accepts, and a dated reason to write.
MeasureaskApollotreg
Usable leads13 of 407 of 406 of 28
Cost per usable lead$0.037$0.27$0.23
Read the comparisonSelf-hosted decisions for ask: Perplexity's open Decider in place of jev on 92 company questionsThe decisions inside hedwigai ask made by Perplexity's open Decider on two NVIDIA RTX 3090s, against jev.
Measureask with jevask with Decider
Questions right, of 928989
Wrong answers12
Read the comparisonClinical decisions: jev and ask on System One, PubMedQA, SciFact and a 370-decision setEdgeEvals' System One board, PubMedQA, SciFact, and a clinical set built to System One's rules.
Measurejevask
Our 370 clinical decisions, right94.6%93.0%
PubMedQA, right (best EdgeEvals scored: 62.2%)77.0%74.4%
Read the comparisonExecutive lookup for US-listed companies: Claude Opus 5.5 and ask on 95 questionsWho runs a public company, and how to email them: a frontier model from memory against hedwigai ask.
MeasureaskClaude Opus 5.5
Questions right, of 955963
Small companies right, of 20139
Read the comparisonRental comps: floor-plan rents and amenities on 14 listings, five assistants comparedRent per floor plan and amenities for US rentals: hedwigai realestate-us against Claude, GPT, Gemini and Perplexity with web search.
Measurerealestate-usBest of four assistants
Floor-plan rents right, of 262623
Amenities claimed the listing doesn't state12129
Read the comparisonCompany screening from SEC filings: jev, Vela 2.0 and ask on 92 yes/no questionsYes-or-no questions about public companies, answered by two decision models alone and by hedwigai ask.
Measureaskjev
Right, of the questions answered98%79%
Wrong answers, of 92219
Read the comparisonDefect localization in industrial images: four frontier models and a trained localizerFour frontier models asked to put a box around what they just described.
MeasureTrained localiserBest of four frontier models
Mean overlap with the defect (IoU, 0 to 1)0.4210.109
Read the comparison