2026-08-17 02:56:01
π Logged-in flows use a persistent browser profile locally or Playwright storageState JSON in CI; stepshots patch amends existing bundles with manually captured steps for one-shot flows
π€ stepshots mcp serves the recording workflow over #MCP on stdio so AI agents can call get_schema, list_tutorials, record, verify and upload
π #Toolport is a local-first #MCP gateway. Every AI client points at one port and shares the same servers, so you set up and authenticate each server once instead of in every app.
#opensource
Six MCP servers here. No key, no account, no OAuth, nothing to install, every tool read-only. The config, complete as it stands:
{"mcpServers":{"ai-crawler-index":{"type":"streamable-http","url":"#MCP
π #LaravelMCP 1.0 introduces searchable tool catalogs: the tools/list payload stays the same size whether a server exposes 10 tools or 100. #Laravel #MCP
π§ Hooks enforce deterministic checks outside the prompt, sessions are distilled into summaries and confidence-scored instincts, and a Memory Vault shares Markdown context across harnesses
π‘οΈ #AgentShield scans prompts, hooks, #MCP configs, permissions, secrets and agent files, treating tβ¦
If you are an agent rather than a person, the same index answers machine-to-machine:
Β· MCP over HTTP at /mcp β JSON-RPC, no key: initialize, then tools/list. lookup_crawler, generate_robots_txt, classify_user_agent, is_verified_crawler_ip.
Β· A2A agent card at /.well-known/agent-card.json β nine skills over message/send.
Β· No client at all? The same tools as keyless GET endpoints under /tools/.
Bulk files, one schema each:
#MCP #A2A
π¨ Inspo β real production websites your coding agent can study before it writes UI #opensource #MCP #webdev #AI
π Hermes Vault: a local-first credential broker and encrypted vault for AI agents. It scans for plaintext secrets, brokers policy-gated access and proves audit integrity.
#opensource #security #MCP
What moved in the crawler index:
Β· Upstream failed: DuckDuckBot (URLError: <urlopen error [Errno 104] Connection reset by peer>)
Β· 4 IP-range lists refreshed at source: Googlebot, Google user-triggered fetchers, Google special-purpose crawlers and 1 more
Β· OAI-SearchBot: 4 prefixes added, 0 removed
Β· Upstream recovered: DuckDuckBot
Every change is in a cursor feed β send the cursor back, get only what is new. Bulk files and the feed:
#AI #opendata #MCP
What moved in the crawler index:
Β· Upstream failed: DuckDuckBot (URLError: <urlopen error [Errno 104] Connection reset by peer>)
Β· 9 document published: ai-crawler-traffic-2026-w36, client-census, crawler-fleet-fold-2026-w36 and 6 more
Β· Upstream recovered: DuckDuckBot
Every change is in a cursor feed β send the cursor back, get only what is new. Bulk files and the feed:
#AI #opendata #MCP
π #HuntProxy is an #opensource web security workbench built for #AI agents β connect it over #MCP
Was that really Googlebot? Check up to 500 addresses at once against the IP ranges the operators publish themselves, do prefix arithmetic on a CIDR, and export the allowlist for nginx, Apache, HAProxy, Cloudflare, ipset or Caddy.
MCP over HTTP at /mcp/netcheck, the same tools as an A2A skill, and as a keyless GET if you have no client at all. Read-only, no signup.
#MCP #A2A #AI #opendata
Does your robots.txt say what you think it says? Paste it and find out: misspelled directives, the noindex line nobody honours, rules that match no path β and the part no other linter does, which of the 150 AI crawlers in this index it actually blocks. RFC 9309 path matching.
MCP at /mcp/robots, A2A at /a2a/robots, no key, no account.
#robotstxt #crawlers #MCP #A2A
229 bots and agents hit one small AI-agent site in 113 hours. Not a vendor list β the host's own log, one page per client: the exact user-agent, how many addresses it came from, the paths it asked for in order, the status codes it got, and what it demanded that does not exist.
Most of these names have no page anywhere on the web: mcpbeat, exaforce-mcprep, Orbit-MCP-Registry-IconResolver. CC0, JSON and CSV.
#crawlers #opendata #AI #MCP