GrantLock Posture — Pre-runtime control plane for agentic AI

Top 5 riskiest servers

View all 128

ServerCritHighMedRisk score

filesystem-mcp483

postgres-mcp265

github-mcp157

linear-mcp046

docs-mcp024

OWASP Agentic Top 10 exposure

LLM01Prompt injection

42LLM02Insecure output handling

28LLM06Sensitive info disclosure

61LLM07Insecure plugin design

34LLM08Excessive agency

Injection & manipulationData exposureAgency & access

Protection coverage

Coverage

96%

137 of 142 checks enabled · +38 Pro detections available

Top 5 disabled checks

Suggested policies

Open Inbox →

87%

Promote insecure-transport check to Block

87% of top-quartile fleets enforce TLS on network MCPs at block-self approval or stricter. You're on Alert.

72%

Enable filesystem path-allowlist enforcement

Most peer fleets block `..` traversal at the scanner level rather than relying on runtime sandboxing.

64%

Suppress legacy template-injection rule

64% of fleets running Jinja-free MCPs suppress this with a 'compensating-control' reason.

41%

Bind tool-count cap per agent

Agents bound to ≤24 tools have 38% lower blast-radius scores in incident retrospectives.

Top detections, last 30 days

Open in Findings →

check.mcp.transport_stdio_sandbox

Server 'filesystem' runs as a local process via stdio.

11 check.crs.logging_capability

Server 'filesystem' does not report a logging capability.

7 check.nhi.static_credential

Environment variable 'GITHUB_TOKEN' appears to contain a static credential passed to the MCP server.

Fleet metrics

24h

Servers scanned (24h)

Top detections · 24h

hits across the fleet

check.mcp.unbound_token

142

check.tool.path_allowlist

check.prompt.injection

check.transport.stdio_preferred

check.tool.unauthorized

check.data.pii_leakage

Posture summary

Worst-day decisions8 days ago · 5× normal

187

Best-day decisions5 days ago · weekend

Median session lengthacross 47 active agents

22m

Coverage driftrule pack vs. prod observations

-2.1%

Overview