Alpha OSINT Prospector

B2B prospecting inside Claude. Describe your ideal customer once, and get back a ranked list of companies that are hiring for the roles you sell into, with funding data, a 0-100 score you can audit, named decision-makers and verified work emails.

It is a Claude Code plugin: a local MCP server with 14 tools, a local SQLite lead store, and 6 skills. It researches. It never sends anything.

Built by Alpha OSINT on our own Apify Actors.

Why another prospecting tool

Most Claude prospecting tools read LinkedIn, one URL at a time, and ask the model to say "High / Medium / Low". That leaves gaps this project fills:

Related MCP server: signal-prospector

The pipeline

ICP  ->  source (job postings)  ->  enrich (firmographics + funding)  ->  emails + decision-makers  ->  score  ->  leads.csv

Related Actors you can chain by hand: LinkedIn Profile API, LinkedIn Posts API, PitchBook Company API, Google Maps Place ID Contact API.

Install

You need uv, an Apify account (a paid plan is recommended, see Cost) and Claude Code.

In Claude Code:

/plugin marketplace add johnisanerd/alphaosint-prospector

/plugin install alphaosint-prospector@alphaosint

When you enable the plugin, Claude Code asks for your Apify API token (Apify Console, Settings, Integrations) and stores it in your system's credential store. It also asks for your Apify plan tier and two spend caps; change those later in /config.

MCP server only (Claude Desktop, Cursor, anything that speaks MCP):

{
  "mcpServers": {
    "alphaosint-prospector": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/johnisanerd/alphaosint-prospector@v0.2.0", "alpha-prospector", "serve"],
      "env": { "APIFY_TOKEN": "your_apify_token" }
    }
  }
}

Use it

Say what you want:

• "Set up my ICP. We sell sales engagement software to US SaaS companies with 50 to 1000 people."

• "Find me 25 leads."

• "Why did the third one score low?"

• "Get me the VP of Sales and an email for the top five."

• "Export the list."

Or run the whole thing without Claude:

uvx --from git+https://github.com/johnisanerd/alphaosint-prospector alpha-prospector run --icp examples/icp.example.json --max-companies 10 --budget 2

Skills

MCP tools

get_config, get_icp, set_icp, estimate_cost, source_companies, enrich_companies, find_contacts, find_emails, score_leads, list_leads, get_lead, export_leads, list_runs, get_spend.

The ICP

One JSON object drives sourcing, filtering and scoring. See examples/icp.example.json.

The score

Unknown data scores zero and the reason says "unknown". An account nobody has enriched cannot outrank one that was checked.

Cost

You pay Apify for the Actor runs, per result. The server estimates each call at list price before it runs, refuses calls over your caps, and starts every run with a platform-side maximum charge.

Defaults: $2.00 per session, $1.00 per call. Change them in the plugin's settings (/config). The standalone CLI reads them from ~/.alphaosint/.env (see .env.example).

Two things to know:

• The email Actor refuses runs whose charge cap is under $0.50, so that stage needs $0.50 of session budget left even when the real cost is a few cents.

• On Apify's free plan that Actor prices its lead and verification add-ons at $0.10 each instead of about half a cent. Use a paid plan, or set leads_per_domain=0.

Limits

• Crunchbase shows round structure and investors publicly, not dates or amounts. Round dates come from the LinkedIn funding summary when LinkedIn has one.

• Job boards match titles loosely. Postings whose title does not contain every word of one of your hiring keywords are ignored and counted, so add each real variant as its own keyword.

• LinkedIn job rows carry no website, so those accounts need enrichment before emails can be found.

• People search returns people whose public profile mentions the company and title, including some who have since left. Prospector keeps only people whose current employer matches the account. Small companies often return nobody; the email step usually finds named people there.

• Indeed job rows currently come without employer details, so Prospector looks each employer up on Indeed's company pages during enrichment to get the website it joins on.

• A catch_all or unknown email verdict means unverified. The export says so.

What it runs, sends and stores

• Runs: a local Python MCP server, started with uv run --locked from the plugin folder. The first start installs the pinned dependencies from uv.lock into the plugin's data directory.

• Sends: only calls to the Apify API (api.apify.com), made with your token, to run the Actors in the table above. Their inputs are your ICP's hiring keywords, locations, company names and website domains, and the job titles you target. Apify bills the runs to your account. Nothing is sent to an Alpha OSINT server, and there is no telemetry.

• Stores: your ICP, the companies, contacts, scores and a spend log in a SQLite file, prospector.db, in the plugin's data directory. Claude Code deletes that directory when you uninstall the plugin. Exports are written where you ask (by default ~/alphaosint-leads.csv).

See PRIVACY.md for the privacy policy.

Privacy and terms

Named contacts and work emails are personal data under GDPR and similar laws; you need a lawful reason to hold and use them. The Actors read public pages. You are responsible for complying with the terms of the sites involved and with anti-spam law where you operate. This project sends no messages.

Development

git clone https://github.com/johnisanerd/alphaosint-prospector && cd alphaosint-prospector

uv sync --locked --extra dev --extra xlsx && uv run pytest -q

Tests use a fake Apify backend and spend nothing.

Credits

Inspired by insaight by Spiros Baxevanakis, which showed how well an MCP server plus skills works for prospect research. Prospector is an independent implementation with a different scope: account sourcing, multi-source enrichment and scoring.

Disclosure: the johnvc/* Actors are built and sold on Apify by the author.

Support: open an issue.

MIT License.