Agent Capability Readiness Network
A readiness layer for AI agents to discover, preflight, and trust callable capabilities. Capability discovery, pre-call checks, and readiness scoring across MCP, CLI, Skills, and APIs — helping agents select capabilities that are available, safe, and reliable at call time.
{
"status": "ready",
"invoke_now_score": 0.91,
"confidence": 0.83,
"valid_for_seconds": 45,
"expected_p95_latency_ms": 3200,
"capacity_state": "ok",
"fallbacks": [
{ "capability_id": "cap_tabula_cli" }
]
}Example preflight response: whether the capability is callable now, the confidence level, and a fallback if it fails.
Background
Tool discovery is well served by registries. What remains is a unified way to evaluate and assure capabilities between discovery and a reliable call.
MCP servers, CLIs, Skills, and APIs are spread across separate registries and repositories, with no unified entry point for discovery and evaluation.
Publishing a tool has a low barrier; quality varies, and listings may carry over-privileged operations or prompt injection risks.
Uptime, latency, rate limits, and recent failures are typically invisible at the moment of the call.
A successful response does not mean the task succeeded, and cross-agent outcomes lack a shared feedback mechanism.
Capabilities
Five subsystems turn fragmented capability sources into discoverable, assessable, and traceable call recommendations.
A unified abstraction over MCP, CLI, Skill, and OpenAPI interfaces, with deduplication and provider resolution.
Learn more →Static checks of permissions, destructive operations, prompt injection, and provider identity before a capability becomes callable.
Learn more →Protocol handshakes, tools/list checks, and read-only probes continuously update each capability's availability.
Learn more →A per-action invoke_now_score, with confidence, evidence level, and an explicit validity window.
Learn more →Capability outcomes reported by agents calibrate scores on task completion rather than HTTP status codes.
Learn more →How it works
Every score is traceable to its source, audit record, probe result, and call feedback.
Public registries, packages, provider submissions
Many interfaces resolved into one capability
Permissions, risk labels, injection scan
Handshake, tools/list, read-only probe
invoke_now_score + confidence + validity
search / preflight for agents
Call feedback recalibrates scores
Outcome feedback flows back into the scoring engine, so evaluations are continuously calibrated against real calls.
Registry preview
Scores are per action, carry a validity window, and always include evidence levels and reason codes.
GitHub · cap_github_repo_tools
Issues, pull requests, code search and repository management over the official MCP server. The most-installed developer capability.
invoke_now_score
0.95
confidence
0.90
Slack (Salesforce) · cap_slack_messaging
Channel messaging, search, and workflow triggers. The default enterprise messaging surface for agents.
invoke_now_score
0.92
confidence
0.86
Google · cap_google_drive_search
File search, content extraction and sharing controls across Workspace documents.
invoke_now_score
0.91
confidence
0.85
Security & trust
Unverified capabilities do not enter recommendations. High-risk actions require explicit confirmation and sufficient evidence.
For developers
Integrate in four steps: search, preflight, call, and report the outcome.
curl -X POST http://101.32.68.30/v1/capabilities/search \
-H 'Authorization: Bearer $API_KEY' \
-H 'Content-Type: application/json' \
-d '{
"query": "extract tables from a PDF into CSV",
"agent_runtime": "claude-code",
"risk_tolerance": "read_only"
}'Early access is open to agent developers, tool providers, and enterprise platform teams.