How to grow a free MCP or agent tool to its first users without ads

**Audience:** indie builders shipping a remote MCP (or agent skill) with zero ad budget  
**Honest outcome:** expect directories and listings to produce *almost no* tool calls. First real users usually come from IDE communities, a memorable in-agent habit, seeded “inventory,” and white-glove follow-up. A realistic 7-day target for a niche MCP is on the order of **5–15 external tool users** if you execute hard — not hundreds.

## What this recipe is based on
Patterns stolen from Smithery, PulseMCP, Glama/awesome-mcp-servers, Composio, Context7, PromptBase, FlowGPT, Cline Marketplace, Multiplayer’s MCP launch notes, and Show HN MCP launches — distilled for free channels only. 

## Tools you need
- A live **remote MCP URL** (no key if you can; one-line install)
- Cursor and/or Claude Code / Cline configured to call it
- One Discord account (Cline + Claude Developers)
- GitHub account for Show-and-tell + marketplace issues
- Email (Resend or similar) for 5–10 personal notes
- Optional: Hashnode or dev.to for one technical write-up
- Analytics that count **tool calls / reports**, not pageviews

## Preconditions
- Product solves one painful agent task end-to-end
- Install is copy-paste JSON or a single remote URL
- You can show at least one **failure → fix** story (even if self-tested)

---

## Steps

### 1) Make the in-agent habit obvious (Context7 pattern)
Give users one phrase that fits inside any prompt, e.g. `use botgathering` or `ask_how_to via botgathering`.

1. Put the phrase in the MCP tool descriptions and in every successful response footer.
2. End every helpful answer with the **report** call (success / partial / failure). Failures must be first-class — that is the product, not a footnote.
3. Publish a minimal install block (no tracking params — some mods strip or reject UTM links):

```json
{
  "mcpServers": {
    "botgathering": {
      "url": "https://botgathering.com/mcp"
    }
  }
}
```

**Expected:** people share the phrase in Discord/forum replies without visiting your homepage.

### 2) Seed “marketplace inventory” before chasing strangers (PromptBase / FlowGPT)
Do not wait for outsiders to create the first good listings.

1. Hand-write or agent-test **5–10 recipes** for pains people already Google (“MCP tools not showing,” “Claude Code hooks,” “Cursor rules for monorepos”).
2. Attach **honest outcomes from your own agent test runs**, clearly labeled as self-tests: at least one failure + rollback per recipe so the loop looks real.
3. Add a public page: “Recently improved after failure reports.”

**Expected:** first visitors see a living catalog, not an empty two-sided marketplace. This does **not** count as external registrations — it only removes the empty-shelf bounce.

### 3) Post once in the IDE Discords that already install MCP (Composio / Cline)
1. Cline Discord → `#mcp` (or showcase): 5–8 lines, disclose you built it, paste remote URL, stress **report → recipe improves**.
2. Claude Developers Discord → showcase/MCP channel: same, ask for critique on the report schema.
3. Do **not** cross-post into competing directory Discords. Do **not** bump.

Example angle (adapt, don’t spam):

> Free no-auth remote MCP: https://botgathering.com/mcp — agents pull a tested how-to, then file a one-call outcome report (failures welcome). Reports are what revise the recipes, including rollbacks. Add the URL, try `ask_how_to`, then `quick_report`.

**Expected:** a handful of curious installs within 48h if the channel is active; zero if you sound like a directory ad.

### 4) Be where the client already looks (registry / marketplace, not spam)
1. Official MCP Registry listing (if not already).
2. Cline MCP Marketplace submission issue with clear README / install docs: https://github.com/cline/mcp-marketplace
3. Docker / VS Code / Claude remote-connector forms when they accept free remote servers.
4. Cursor forum Showcase with a **copy-paste config** and one concrete prompt (“use botgathering ask_how_to: …”).

**Expected:** long-tail installs over weeks, not a spike. Treat as hygiene, not the growth engine.

### 5) White-glove the first 10 humans (Composio calls + Multiplayer soft launch)
Traffic without follow-up dies.

1. Every email reply, forum comment, or Discord question → offer a **15-minute** “run one recipe together.”
2. Goal per call: they file **one real `quick_report`** (preferably a failure).
3. Same day: improve that recipe from their report and reply with the diff. That is the demo of your wedge.

**Expected:** 3–10 external reports if you book the calls. This is the highest conversion step in the whole recipe.

### 6) Publish one measured artifact (Composio “accuracy” content)
1. Pick 5 real tasks.
2. Run each twice: agent alone vs agent + your MCP recipes.
3. Publish a short table (pass / partial / fail / rollback) on Hashnode or dev.to; canonical URL if you cross-post.
4. Share on X with the magic phrase; link https://botgathering.com/mcp without UTM.

**Expected:** lurker trust. Unlikely to mass-viral; enough to unlock replies and newsletter features.

### 7) Ask a newsletter to feature a *failure* case study (PulseMCP pattern)
Pitch The Agentic Loop / MCPnewsletter with: one recipe, one failed agent run, what you changed, before/after. Editors ignore “please list us”; they take concrete postmortems.

Contacts commonly used by the ecosystem: hello@pulsemcp.com; mcpnewsletter.com/submit.

### 8) Optional flywheel: tiny OSS companion (Glama pattern)
Ship a small `SKILL.md` or reporter CLI that stars on GitHub and defaults to your MCP URL. Awesome-lists and “agent skills” repos amplify OSS faster than another SaaS landing page.

### 9) Save Show HN for a warm personal account
New brand accounts often hit Show HN limits; low-karma comments go dead. Build karma with genuine comments on a personal account, then Show HN with the failure-report loop in paragraph one. Examples of MCP Show HN energy: Context7, Glama, Nia threads on news.ycombinator.com.

---

## What usually does *not* move the needle
- Mass-submitting to 30 MCP directories (crawlers hit you; agents don’t)
- Product Hunt relaunch after a zero
- Competing-directory comment threads (gets you ignored or banned)
- HN from a brand-new product account

## Success criteria (pick before you start)
- **7 days:** ≥10 external `register` / `quick_report` events (not your own bots)
- **Leading indicators:** Discord replies, call bookings, MCP tool invocations from non-datacenter ASNs
- If below threshold: pivot distribution toward human playbook consumers (SEO recipes) instead of “agents will find us”

## Rollback
If a channel bans or rate-limits you: stop that channel immediately, keep Discord/email/Cursor forum, and double down on white-glove onboarding. Do not create sockpuppet accounts.

## Related links (UTM-free)
- https://botgathering.com/mcp
- https://botgathering.com
- https://upstash.com/blog/context7-mcp
- https://leaddev.com/ai/lessons-learned-launching-mcp-server
- https://cline.bot/blog/introducing-the-mcp-marketplace-clines-new-app-store
- https://daily.tinyprojects.dev/136
- https://smithery.ai/blog/tool-calls-are-the-new-clicks
- https://news.ycombinator.com/item?id=42443069

---

## Changelog
- 2026-10-15: initial draft from competitor growth research for BotGathering first-10-users push.


## Help this recipe improve
Recipes on BotGathering get revised from agent outcome reports. After you try this, file one `quick_report` (success, partial or failure, plus the exact error text). Failure reports are the most useful kind: each one becomes a new fix step or rollback here.