Getting your first customers is rarely a writing problem. It is a finding problem.
Somewhere today a person typed "is there a CRM that does not need a sales team," or posted on Bluesky that they are finally ditching their email tool, or left a YouTube comment spelling out the one feature a competitor never shipped. Those are your customers, mid-sentence, telling you what they want. The catch is they are spread across five platforms and a few thousand communities, and you are one founder with one set of eyes.
So hand the reading to an agent. Below is how to set one up that surfaces those people and drafts the reply, without becoming the thing everyone mutes.
Start before the send button
Most "AI agent for marketing" guides open at the end of the funnel. The agent writes the post, fires the DM, updates the CRM. That is fast, but it is fast guessing. If the agent does not know who is in pain right now, it is just producing volume.
The piece that actually changes the math sits earlier: one tool that turns a plain idea into ranked buyer intent. Who is asking publicly, in their own words, sorted by how close they are to paying. Get that right and the reply almost writes itself.
That tool is the Gorilla MCP server. It is an open Model Context Protocol server, so any agent that speaks MCP (Claude, Cursor, your own) can call it. One call fans out across Reddit, X, Bluesky, LinkedIn, and YouTube at once, then scores every post by buying intent. What comes back is real links to real people, not a hallucinated list.
Get a key
Sign up at usegorilla.app, then make a key under API Keys at platform.usegorilla.app. It starts with grla_. Billing is per run, not per seat. A full search is a flat $5, and there is no subscription to forget about.
Wire it into your agent
Point your agent at Gorilla. In Claude Code or Cursor that is a few lines of MCP config:
{
"mcpServers": {
"gorilla": {
"command": "npx",
"args": ["-y", "@usegorilla/mcp"],
"env": { "GORILLA_API_KEY": "grla_your_key_here" }
}
}
}
Nothing to host. npx grabs the package, the agent reads the available tools, and you are wired up. The same block works anywhere MCP is supported.
Let it run the loop
Now you just talk to it. Tell the agent, "find people who need a project management tool for agencies." It calls leads.find, waits out the run, and comes back with scored leads grouped by platform. Every lead carries a link, the intent signals it matched, and a score from 0 to 1.
Then it keeps moving. Pass the run to outreach.plan and it lays out a Week-1 plan, how many replies per platform and in what order. Pass it a single lead with outreach.draft and it writes a reply that fits where the person actually posted. A Reddit comment is not a LinkedIn note, and the drafter treats them differently. The whole thing, find to first draft, happens in one conversation. You never open a tab.
What it can reach for
Once Gorilla is connected, the agent has a short, sharp set of tools:
leads.findruns the full pipeline and returns scored leads.leads.searchhits one platform with your exact queries when you want to go deeper.idea.refineandidea.expandturn a one-line idea into keywords, pain points, and competitors before it searches.outreach.draftwrites a per-lead reply.outreach.planbuilds the weekly cadence.runs.getandruns.listre-open past hunts without burning another credit.
The part that stays yours
The agent finds the person and writes the first draft. Whether it gets sent is on you, and it should be.
The quickest way to torch your first hundred prospects is to let a bot spray generic replies into threads where everyone can tell. Let the agent do the reading, the ranking, and the rough draft. Keep the edit and the send. One reply that clearly read the post beats fifty that clearly did not.
Try it
You do not need a list or an audience. You need to reach the people already asking before someone else does. Set the agent up once and let it read the internet for you every morning.
Grab a key at usegorilla.app and point your agent at it.