CTF OSINT - MCP Server Integration
OSINT is browser-heavy: reverse-image search, map/Street View exploration,
social-media profiles, archived pages, and JS-rendered sites that curl cannot
read. Playwright MCP drives a real browser from the agent so it can navigate,
screenshot, read rendered content, and step through interactive search flows.
Table of Contents
- Setup
- Rendered-Page Reading & Screenshots
- Reverse-Image & Map Search Flows
- Social Media & Archived Pages
- When to Use Playwright MCP vs curl/API
Full tool list and the general browser workflow live in ../ctf-web/tools-mcp.md.
Setup
Repository: https://github.com/microsoft/playwright-mcp (Apache-2.0, official
Microsoft, tested v0.0.78+).
npx @playwright/mcp@latest # most MCP clients auto-launch this
# The browser binary is a separate download; without it every call errors with
# 'Browser "chrome-for-testing" is not installed':
npx @playwright/mcp@latest install-browser chrome-for-testingSnapshots, console logs and non-text response bodies are written to
.playwright-mcp/ in the working directory and returned as file paths — read the
file, and keep that directory out of version control.
Rendered-Page Reading & Screenshots
Modern OSINT targets render content with JavaScript — a raw fetch returns an empty shell. Playwright reads the page after JS runs.
browser_navigate "https://example.com/profile/target"
browser_snapshot # rendered text/links/roles as an accessibility tree
browser_take_screenshot # capture the visual for geolocation clues
browser_evaluate "() => document.body.innerText" # full rendered text dumpKey insight: For any site that shows a blank page or a loading spinner under
curl, Playwright MCP is the fix — browser_snapshot returns the content the way
a human sees it, after client-side rendering, so names, timestamps, and hidden
metadata become readable.
Reverse-Image & Map Search Flows
Reverse-image search and map exploration are multi-step interactive flows the agent can script end to end.
# Reverse image search
browser_navigate "https://images.google.com"
browser_click <camera / "Search by image" ref from snapshot>
browser_type <url field ref> "https://example.com/clue.jpg"
browser_press_key Enter
browser_snapshot # read the visual-match results + candidate locations
# Map / Street View narrowing (cross-ref geolocation-and-media.md)
browser_navigate "https://www.google.com/maps/@<lat>,<lon>,3a,75y/data=..."
browser_take_screenshot # compare panorama landmarks against the target photoKey insight: Geolocation challenges are iterative — search, read candidates, refine, re-search. Playwright MCP lets the agent run that loop autonomously: submit the image, read the top matches from the snapshot, then pivot to maps to confirm landmarks. Pair with the manual techniques in geolocation-and-media.md.
Social Media & Archived Pages
browser_navigate "https://web.archive.org/web/*/example.com"
browser_snapshot # list all snapshots; open one to read content deleted from the live site
browser_navigate "https://example.com/@target"
browser_evaluate "() => document.body.innerText" # bio, post timestamps, geotags
browser_network_requests static=false filter="/api/" # find the profile JSON endpoint
browser_network_request index=<n> part="response-body" # read the JSON itselfKey insight: a social site's profile JSON API is where the real intelligence
is — IDs, exact timestamps, coordinates that the rendered HTML rounds off or
omits. But browser_network_requests only lists the endpoints; the structured
data comes from browser_network_request part="response-body". Take both steps,
then pivot to username/metadata mining (see social-media.md
and web-and-dns.md).
When to Use Playwright MCP vs curl/API
| Situation | Use |
|---|---|
| JS-rendered page, infinite scroll, or login wall | Playwright MCP |
| Interactive reverse-image / map search flow | Playwright MCP |
| Reading archived (Wayback) snapshots visually | Playwright MCP |
| Bulk WHOIS / DNS / cert transparency lookups | CLI/API (see web-and-dns.md) |
| Structured API with a documented endpoint | requests/API — faster, no browser |
Key insight: Use Playwright MCP when the intelligence is locked behind client-side rendering or an interactive UI. For structured lookups (DNS, WHOIS, cert transparency, platform APIs) the direct CLI/API path stays faster and more scriptable.