Skip to content
FlyTrap-QAPublic

About

FlyTrap MCP Server.

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

@flytrap-qa/mcp

MCP server for FlyTrap.

FlyTrap uses your Android or iOS app on real devices the way a new user would: signing in, going through onboarding, searching, hitting the wrong password screen. In about 30 minutes you get a list of where it breaks. This server puts that list in the coding agent you already use (Claude, Cursor, VS Code, Devin, Codex, Windsurf, Zed), with the screenshot, UI XML, logcat and failed network calls for every bug, so the agent can go and fix them.

It can also read the screen graph, the hosts your build talks to, detected SDKs and permissions, and what changed between two builds. It only reads results. It does not run apps, decompile binaries or touch your source tree.

There are two ways to connect: the hosted server (nothing to install, you sign in with your FlyTrap account) or this npm package running locally with an API key.

Hosted server

URL: https://api.flytrap.dev/mcp (Streamable HTTP, OAuth)

Add it as a remote MCP server. Your agent opens a browser, you sign in to FlyTrap and click Allow. Read access is always on; tick Also allow starting crawls only if you want the agent to queue crawls. The connection shows up in the dashboard under Settings > API keys, and you can revoke it there.

  • Claude (claude.ai, Claude Desktop): Settings > Connectors > Add custom connector, paste the URL.
  • Claude Code: claude mcp add --transport http flytrap https://api.flytrap.dev/mcp, then run /mcp, pick flytrap and choose Authenticate.
  • Cursor (~/.cursor/mcp.json): { "mcpServers": { "flytrap": { "url": "https://api.flytrap.dev/mcp" } } }
  • VS Code (.vscode/mcp.json): { "servers": { "flytrap": { "type": "http", "url": "https://api.flytrap.dev/mcp" } } }
  • Devin: Customize > MCPs > Add custom MCP, HTTP transport, paste the URL, authentication OAuth, then Connect.

If your client can't do the browser sign-in, send an API key instead: Authorization: Bearer ft_live_....

Local package

  1. Sign in at dashboard.flytrap.dev, open Settings > API keys and create a key. Leave Allow starting crawls off unless you want the agent to be able to queue new crawls.
  2. Add the server to your agent:
{
  "mcpServers": {
    "flytrap": {
      "command": "npx",
      "args": ["-y", "@flytrap-qa/mcp@latest"],
      "env": {
        "FLYTRAP_API_KEY": "ft_live_..."
      }
    }
  }
}

Claude Code: claude mcp add flytrap -e FLYTRAP_API_KEY=ft_live_... -- npx -y @flytrap-qa/mcp@latest

Codex CLI (~/.codex/config.toml):

[mcp_servers.flytrap]
command = "npx"
args = ["-y", "@flytrap-qa/mcp@latest"]
env = { FLYTRAP_API_KEY = "ft_live_..." }

What to ask

  • "Where does my app break? Show me the evidence."
  • "Which third-party hosts does build 1184 talk to before the consent screen?"
  • "What changed between build 1180 and 1184: SDKs, permissions, bugs?"
  • "Which SDKs does com.booking ship on Android?" (public catalog, no key needed)

Without FLYTRAP_API_KEY the public catalog tools still work.

Tools

Your projects (need a sign-in on the hosted server, or FLYTRAP_API_KEY locally):

Tool What it returns
whoami The account the key belongs to
list_projects Projects the key owner can read, with the selected crawl run
get_project One project plus its live crawl status
list_builds Builds of a project with binary SHA-256 and provenance
get_build_summary Version, size, SDK/permission counts, bug and test counts, screens, network and performance totals
get_sdks Detected SDKs with confidence and evidence sources, declared permissions, deep links
list_bugs Bugs from the selected run, sorted by severity
get_bug_evidence Scenario, failed step, screenshots and XML (signed URLs), logcat error window, failed network calls, performance peaks, a ready ai_prompt
get_screen_graph Screens reached and the tap edges between them
get_network_hosts Hosts contacted with call counts, bytes and error rates
search_apps Describe an app in plain language, get the crawled public apps whose observed features cover it, with matched and missing requirements
get_ux_benchmark Your build's features next to the same feature in competitor and same-genre apps, with screenshots
list_competitors Competitors pinned to a project's UX benchmark
find_competitors Search crawled public apps and Google Play for apps to pin
add_competitor Pin a crawled app, or create a store project from a Google Play package and pin it (crawl scope, confirm: true to create)
remove_competitor Unpin a competitor (crawl scope)
compare_builds SDK, permission, deep link, host, bug and screen diff between two builds
start_crawl Queues a new crawl of the current build. Needs the crawl scope and confirm: true

Public catalog (no key): lookup_app, get_app_profile, lookup_sdk, lookup_permission.

Agent skill

skills/flytrap/SKILL.md tells an agent when to reach for which tool and how to cite evidence. Copy it into your project's skills folder (for example .claude/skills/flytrap/SKILL.md or .cursor/rules/) or point your agent at it.

Configuration

The only setting is FLYTRAP_API_KEY: your personal key from Settings > API keys (ft_live_...). It is optional for the public catalog tools. The server always talks to https://api.flytrap.dev.

Security

  • The key is sent only to https://api.flytrap.dev as a Bearer header, over https. It is never written to disk or logged.
  • Keys are scoped. A read key cannot start crawls, create keys or change billing, even if the agent asks. Revoke keys any time in the dashboard.
  • Keys inherit the project permissions of the user who created them. A key never sees another account's private projects.
  • Tool output is bounded (log windows, call caps, 4 MiB per response) and marked as untrusted app data, since bug titles, log lines and URLs originate from the crawled app and its traffic.
  • start_crawl, add_competitor and remove_competitor are the only tools that change anything; they need a key created with the crawl scope. Queuing a crawl (start_crawl, or add_competitor with a Google Play package) also requires an explicit confirm: true.

REST instead of MCP

Every tool maps to a REST endpoint, so an agent that can run curl can do the same without this package:

curl -H "Authorization: Bearer $FLYTRAP_API_KEY" https://api.flytrap.dev/project/projects
curl -H "Authorization: Bearer $FLYTRAP_API_KEY" https://api.flytrap.dev/project/projects/1248/bugs
curl -H "Authorization: Bearer $FLYTRAP_API_KEY" https://api.flytrap.dev/project/projects/1248/bugs/BUG-102/evidence

API reference: https://api.flytrap.dev/docs

Development

npm install
npm test
FLYTRAP_API_KEY=ft_live_... npm run inspect   # MCP Inspector UI

Requires Node 20 or newer. Published from GitHub Actions on tags (v*) with npm provenance.

About

FlyTrap MCP Server.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages