DEV Community: AgentoolRank

What submitting to 101 directories taught us about free listings

AgentoolRank — Wed, 07 Oct 2026 13:17:25 +0000

81 of 101 directories offered a free option. Budget time for inbox steps and captchas, and read the free tier's conditions before filling the form.

Our own submission notes cover 101 sites across three products, tested September 29 through October 2, 2026, and last verified October 2.

I'm Jason T., maker of AgentoolRank, our directory of open-source AI agent tools ranked by live GitHub activity. We also run two other small products.

What the free option actually bought us

We worked by hand and with browser automation, with one submission per site per product and no captcha-solving services. At least one of our products went through each site's submission flow.

The headline needs three qualifications:

56 sites put free submissions in a review queue.
30 sites required an inbox step. Verification links, magic links, and codes needed someone available.
66 sites ended as submitted or already listed.

Here is the full outcome breakdown:

Outcome	Sites
Submitted or already listed	66
Opened, then skipped for rules or fit	19
Failed on the first try, queued for retry	9
Stopped at a visible captcha	5
Stopped because the free tier required a badge we declined	2

Keep "skipped" in your log.

Budget for handling, then budget for waiting

Separate your working time from the directory's waiting time.

For hands-on work, reserve attention for inbox checks and account access. Besides the 30 sites requiring email interaction, 19 offered or required Google login, and 8 had a captcha a person had to complete. These categories can overlap. The 8 captcha sites describe flow requirements; the 5 captcha-stopped outcomes describe where our attempts ended.

Inbox steps alone mean a fully unattended agent will stall on about a third of these sites.

For elapsed waiting time, the examples were concrete:

Directory	Observed free submission wait or condition
SaaSHub	Free no-login form can take up to 32 days
TinyLaunch	Scheduled our free launch for November 2, 2026
Viesearch	Showed 1,200+ queued submissions and stated an 82% free-tier rejection rate

Also reserve a follow-up pass. Only 75 sites showed a clear confirmation, such as FutureTools' "Tool Submitted!" For the others, the last click was not enough to establish what happened. Check later before submitting again.

A shortlist ordered by conditions you can accept

We read each free tier's conditions before filling the form. Here is what those conditions looked like.

For sites that fit, this is the order I would use to triage the observed options.

Priority	Candidate	When to consider it	Friction to account for
Start here	FutureTools	Its rules fit your tool and you want a clear submission signal	Dismiss the newsletter popup with "No thanks" before using the form; look for "Tool Submitted!"
Start here if waiting is acceptable	SaaSHub	Its rules fit and the free no-login route suits you	Up to 32 days; claim the listing before closing the browser or lose management access
Schedule separately	TinyLaunch	The offered launch date works for your plans	Our free launch was scheduled for November 2, 2026
Conditional for AI agent tools	AI Agents Directory	Its eligibility rules fit and a free `nofollow` link is acceptable	Free links are `nofollow`; dofollow starts at $49
Defer if queue risk matters	Viesearch	You accept its stated queue and rejection conditions	Check spam, then click "Join the Waiting List" and "Wait in Line"
Ongoing badge condition	Findly.tools	Free listing tied to a footer badge	Our notes say removing the badge later can lead to delisting

This order changes with your constraints.

Across the dataset, 14 sites required a badge, 12 required a backlink, and 1 required an X post.

Of the 39 sites that showed a price, the median entry price was $12.

Copy this submission log before opening forms

Use a separate row for each site and product. Keep a log of the date, result, and what blocked you so you can avoid duplicate submissions and follow up.

Site:
Product:
Submission date:

Fit decision:
Free-tier conditions:
Badge or reciprocal-link commitment:

Attempt result:
Confirmation text or evidence:
Blocker:
Inbox or account action still needed:

Expected review or launch date:
Next follow-up action:
Follow-up date:

Live listing URL:
Observed link rel:
Management access claimed:
Continuing requirements:

Use the same outcome labels consistently: submitted or already listed, skipped, retry queued, captcha-stopped, or badge-declined. Add a note when confirmation is uncertain.

Make follow-up instructions specific enough to act on:

SaaSHub: Claim management access before ending the browser session. Record the review wait separately.
Viesearch: Check spam and complete both waiting-list actions.
Findly.tools: The site can delist you if you remove the footer badge later.
Unclear confirmation: Check for the listing later before retrying.

For a failed first attempt, write down the blocker. Otherwise, the next session starts by rediscovering the same problem.

Verify the link before counting it

We inspected link rel attributes on 30 sites, usually when something looked off. This was not a random sample.

Recorded link category	Sites
`nofollow`	23
Dofollow	6
`ugc`	1

AlternativeTo listings were nofollow, as were LaunchBoosts free-plan links. make.rs profile links used rel="ugc nofollow".

We only count a dofollow link after inspecting the live listing.

Stars measure attention, downloads measure use

For open-source tools, keep distribution signals separate from evidence of use, too. Stars measure attention; package downloads offer a usage signal, not a user count.

The downloads comparison counts npm and PyPI downloads over 30 days using the npm downloads API and pypistats. A package counts only when its registry page links back to the tool's own repository. Downloads include CI and mirrors.

Spend the next session on fit and follow-up

We saw very little referral traffic so far. We make no promise of traffic, publication or ranking gains.

Check conditions before filling forms, reserve time for inbox and captcha steps, and keep a dated log of each site's result and blocker.

Use our free comparison of where to list an AI agent tool to choose where to spend that submission time.

If a site's rules changed, tell us and we update the data.

Our Submit Kit ranks these directories for your product type, with form gotchas and a don't-submit list; the top 10 are free over MCP, and the full list is $29 one-time.

The free top 10 for each product type is also on the web: AI tool, MCP server, developer tool, SaaS.

Jason T.

Disclosure: Submit Kit is our product.

Where to list an MCP server or AI agent tool: free vs paid (checked Oct 2026)

AgentoolRank — Sat, 03 Oct 2026 00:38:07 +0000

Disclosure: I build AgentoolRank, one of the directories in the table below. Prices and review times were checked on each site's submit page on 2026-10-01; they change, so check the links before you decide. The up-to-date version of this table lives at agentoolrank.com/where-to-list, which now also has a filterable list of 101 general directories our products went through (free options, badge or backlink conditions, login and captcha).

This week I shipped a public MCP server (https://agentoolrank.com/api/mcp, which can search 593 open-source agent tools, pull GitHub stats and alternatives) and went through the places an MCP or agent-tool builder would reasonably list it. I published to the MCP Registry, submitted to mcpservers.org and PeerPush's free queue, and checked the submit pages of the rest. Here's what each asks for and what's free.

The short version

Almost everything has a free route if you can wait. Paid tiers mostly buy speed, placement or a dofollow link.
For an MCP server, publish to the official MCP Registry first. It's free, it's immediate, and other directories read from it.
mcpservers.org is free (up to 2 weeks). mcp.so's web form is paid ($39); there's a free GitHub-issue route.

The table

Directory	Free option	Paid options	Review time
Official MCP Registry	Yes, via `mcp-publisher` CLI	None	Immediate after domain/GitHub verification
mcpservers.org	Yes	$39 (≤24h, badge, dofollow)	Free: up to 2 weeks
mcp.so	Not on the web form (GitHub issue route)	$39 one-time	Paid: instant
PeerPush	Yes, free queue (we were quoted ~70 days)	$39 · $89 · $229	Paid: instant
AI Agents List	Eligibility check only	$29 lifetime · $49 launch boost	48h / 24h
AgentoolRank (mine)	Yes, queued review	$9 · $19 · $49	Free: queue order

Publishing to the official MCP Registry (the one worth doing first)

This took about 10 minutes. If you own a domain, DNS verification lets you use a com.yourdomain/... name instead of io.github.you/...:

Add a server.json describing your server (name, description, version, and a remotes entry with your Streamable HTTP URL).
Generate an Ed25519 key, put the public key in a TXT record on your domain (v=MCPv1; k=ed25519; p=...).
mcp-publisher login dns --domain yourdomain.com --private-key <hex> then mcp-publisher publish.

Two things that tripped me up:

The login token expires quickly. If publish fails with an auth error a while after logging in, just run login again.
The name in server.json must match the namespace you verified (com.agentoolrank/... for agentoolrank.com).

What I'd do again

MCP Registry → 2. mcpservers.org (free) → 3. one general directory where builders browse (PeerPush's free queue) → 4. stop, and spend the rest of the time on content people actually search for.

Paying to jump queues only made sense to me if launch timing mattered. For a side project, the free queues were fine.

If you've found another MCP or agent-tool directory that's worth the time, I'd like to hear about it. I'll add it to the live table after checking its submit page.

Stars lie: 669 open-source AI agent repos ranked by what they actually do

AgentoolRank — Thu, 01 Oct 2026 06:29:37 +0000

Disclosure: I build AgentoolRank, the directory these numbers come from. The data is refreshed daily from the GitHub API and the full tables are free to reuse (CC BY 4.0).

I kept picking AI agent frameworks the way most of us do: sort by stars, skim the README, pip install. Then, weeks later, I'd find the repo hadn't merged anything in months.

So I started tracking 669 open-source tools for building, running and evaluating AI agents — frameworks, coding agents, MCP servers, RAG and memory layers, eval tools — and ranking them by activity instead of stars. Here's what that looks like.

1. A third of the ecosystem has gone quiet

213 of 669 tools (32%) have had no commit on their default branch in six months or more. That includes 68 projects with 5,000+ stars:

Project	Stars	Last commit
MetaGPT	70.7k	Jan 2026
gpt-engineer	55.1k	Nov 2024
Grok-1	52.2k	Mar 2024
Quivr	39.6k	Jun 2025
FastChat	39.6k	Jun 2025
Langchain-Chatchat	38.7k	Nov 2025
AgentGPT	36.3k	Apr 2025

Stars are a record of past attention. They say very little about whether issues get answered or whether the thing works with this month's model APIs.

2. The fastest growers add 20k+ stars a month

Star pace over our daily snapshots, scaled to 30 days:

Project	Stars	+30 days
Skills	179k	+26.3k
LangChain	147k	+23.5k
ECC	270k	+22.5k
hermes-agent	250k	+20.9k
DeepSeek Harness	241k	+20.1k
MarkItDown	188k	+15.3k
Firecrawl	187k	+14.1k

3. Some repos commit hundreds of times a day

Commits on the default branch in the last 90 days (I double-checked the top ones against the GitHub REST API):

hermes-agent — 32,240
OpenHuman — 21,133
elizaOS — 19,953
DeepSeek Harness — 19,798
LiteLLM — 13,153

That's 150–360 commits a day. A good chunk of that is almost certainly agents committing to their own repos — which is its own interesting signal about where this ecosystem is heading.

4. Where the growth is, by category

Category	Tools	Stars added / 30 days
Agent frameworks	329	+380k
Tool integration & infrastructure	128	+165k
Memory & knowledge	127	+150k
Coding agents	42	+92k
Enterprise agent platforms	57	+75k
Observability & evaluation	70	+68k

MCP-first tools are now a category of their own: 29 projects, about 580k stars combined.

5. How the ranking works

Four signals, refreshed daily from the GitHub API:

Star pace — stars gained per 30 days, from daily snapshots (not the lifetime total).
Commits in the last 90 days on the default branch.
Releases in the last 6 months.
Issue response — median time to close recent issues.

They're combined into a percentile score. None of them is perfect on its own (commit counts reward bots; stars reward launch hype), which is why the site shows all four side by side, plus an alternatives page and head-to-head comparisons for every tool.

6. Query it from your own agent

If you use Claude, Cursor or any MCP client, you can point it at the data directly:

{
  "mcpServers": {
    "agentoolrank": { "url": "https://agentoolrank.com/api/mcp" }
  }
}

Tools: search_tools, get_tool, get_alternatives — and submit_tool if you want to list something you've built. There's also a plain JSON API; details at agentoolrank.com/agents.

The full, daily-updated tables are at agentoolrank.com/report. What would you weight more heavily than stars when picking a framework — release cadence, issue response, something else?