Is your MCP server actually installable by an agent?

A server that works on your laptop is not the same thing as a server an agent can find, install and call on its own. These are the 12 checks that decide whether that happens — the ones we had to pass to ship TAPAC as npx -y @tapacapi/mcp.

Tick what you already have. The score is computed in your browser from your own answers; nothing is sent anywhere.

Score

1. Discovery — can the agent find you at all?

A stable machine-readable manifest exists at a fixed URL (llms.txt, SKILL.md, AGENTS.md or plugin.json).Agents are pointed at URLs, not at your marketing page. One file, one URL, no auth. The repo carries the mcp-server / mcp topic and a one-line description that says what it returns.Directories index by topic; the description is what an agent quotes back. Installable from a public registry (npm / PyPI), not only from a git clone.Test: npm view <pkg> version returns a version. The install block is copy-pasteable JSON for the main clients (Claude, Cursor, Codex, VS Code).A human should not have to read source to wire it up.

2. Install & transport — does it start with nothing?

npx -y <pkg> (or uvx) runs the server over stdio with no prior state.No build step, no git clone, no npm install first. It boots and answers tools/list without credentials.Missing key = informative onboarding response, not a stack trace. Otherwise the agent gives up on step one. The transport is stated explicitly — stdio and/or a hosted HTTP endpoint."It's an MCP server" is not a transport.

3. Tool surface — can the agent use it without guessing?

Every tool description states what it returns and what the required arguments are.The description is the prompt. Vague descriptions produce invented arguments. Inputs are constrained by a schema (types, enums, min/max), not free-form strings.Enums stop the agent from passing industry: "SaaS-ish". Failures return a structured, actionable error instead of an empty result."No contacts matched — widen company_size" beats []. There is a status/health tool reporting version, credential state and upstream reachability.This is the first thing a debugging agent will call.

4. Trust — can an agent recommend you?

Pricing is public and per use, with no sales call and no seat licence in the way.A price an agent can quote is a price a buyer can approve. Every factual claim in the README has a name next to it (source, or a command you can paste).Unsourced numbers are what makes reviewers close the tab.

Test the package, not the README

npx -y @tapacapi/mcp        # boots, answers tools/list, asks for a key instead of guessing
npx -y @tapacapi/mcp --help # version + install notes

Running the same three commands against your own package is the whole audit: does it start clean, does it list tools, does it behave without a key.

How TAPAC scores on its own list

Check	TAPAC
Manifest	SKILL.md, AGENTS.md, plugin.json, llms.txt
Registry	npm @tapacapi/mcp, PyPI tapac-sdk
Transport	stdio via npx/uvx + hosted endpoint
Tools	tapac_find_contacts, tapac_status
No key	returns onboarding, not errors
Pricing	$0.10–0.50 per contact, 100 searches free — tapacapi.com/get-key

Do the same audit for the data layer

Readiness gets you installed. Verified contacts are what keeps the loop worth running: lead cost & bounce-risk calculator · TAPAC vs Apollo / ZoomInfo / Hunter / Lusha.

Get a free API key Read the repo