{"name": "In-band Sleeper Sentinel", "provider": "TextForAgents LLC", "version": "1.0.2", "what_it_does": "Sits in-band, inside your production client's own MCP session \u2014 same client identity, same per-client counter \u2014 and diffs every mid-session tools/list refresh, prompt response, and tool result against the approved baseline before it reaches your model's context. Changes are quarantined with the baseline substituted and raised as a re-approval event; results are screened jointly with the server's descriptions and prompts (the assembled view, new-ingredient rule); during idle the package-side sentinel advances per-client counters on harmless probes, so a gated payload fires on a probe instead of on your credential-bearing task.", "prices": {"per_server": "$0.25/server/mo (x402) \u2014 in-band sentinel watching one server", "fleet": "$19/mo per operator \u2014 up to 50 servers, fleet verdict/event surface", "registry": "$99/mo per platform \u2014 watched-server scope sized at onboarding and recorded", "trial": "first month free on the first 5 servers, one time per operator"}, "pay_to": "0x7a482F3B58780519880a6A13F983E958f2Ed6A61", "how_to_start": "POST /register {\"operator\": \"your-name\"} -> token (shown once); POST /trial with the Bearer token claims the one-time grant (first month, first 5 servers); POST /servers registers watched servers within your cap; POST /subscribe {\"plan\": \"per_server\"|\"fleet\"|\"registry\", \"servers\": N} is x402-gated at $0.25/server, $19, or $99; POST /inspect runs the sentinel over a session transcript; the embeddable package is at GET /package/.", "endpoints": {"register": "POST /register", "trial": "POST /trial", "servers": "POST /servers, GET /servers", "subscribe": "POST /subscribe", "inspect": "POST /inspect", "verdicts": "GET /verdicts", "events": "GET /events", "account": "GET /account", "revoke": "POST /tokens/revoke", "mcp": "POST /mcp", "package": "GET /package/in_band_sleeper_sentinel-1.0.2.tar.gz", "openapi": "GET /openapi.json", "llms": "GET /llms.txt", "x402_discovery": "GET /.well-known/x402", "health": "GET /health"}, "limits": ["Admission-before-calls contract: onboarding completes baseline capture before tool calls begin; a surface first seen after calls have begun surfaces an admission warning, never a silent approved baseline.", "The rule screen catches category-complete instruction shapes and does not claim encoded/obfuscated instructions; the diff leg (content-independent) still catches every metadata change.", "The stdio proxy pairs one response line per request; servers emitting extra unsolicited lines are a documented v1 boundary (every line actually read is still inspected fail-closed).", "Verdicts are per-(server, client), naming server, client, and call count; no surface issues a surface-wide clean bill.", "Screening is linear in result size with a documented cap: results over 8M chars of text are quarantined under the large-result policy instead of stalling a session.", "Proven in the lab (LABREPORT-077): 59/59 tests; gated turns quarantined 100% in-process and over stdio; benign false positives 0/8 including HARD fixtures."], "support": "support@textforagents.com"}