Skillget

kontext

mcp
Security Audit
Fail
Health Pass
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 210 GitHub stars
Code Fail
  • rm -rf — Recursive force deletion command in internal/cedareval/testdata/portable/v1/authorization-v1.json
Permissions Pass
  • Permissions — No dangerous permissions requested
Purpose
This CLI tool acts as a secure wrapper for AI coding agents. It manages identity and injects short-lived, scoped credentials into agent sessions while logging tool calls for auditing, eliminating the need to store static API keys in local `.env` files.

Security Assessment
Overall risk: Low. The tool is explicitly designed to handle sensitive data (exchanging secrets and authenticating via OIDC) and does so by storing tokens in the system's secure keyring rather than plaintext. It executes shell commands safely to launch and wrap underlying AI agents. It requires network requests to communicate with its backend for credential resolution and dashboard streaming. The automated code scan found no dangerous patterns, hardcoded secrets, or excessive permission requests. The architecture relies on a cloud backend for token exchange, meaning you must trust their infrastructure to handle your service credentials.

Quality Assessment
The project demonstrates strong maintenance and health. It utilizes the highly permissive MIT license. The codebase is active, with repository updates pushed as recently as today. It has accumulated 50 GitHub stars, indicating a modest but growing level of community trust. Additionally, the project provides comprehensive documentation, installation guides, and a Discord server for user support.

Verdict
Safe to use.
SUMMARY

Runtime Security for tool-using AI agents, with local policies, pre-action enforcement, and forensic audit trails.

README.md
Kontext banner

Runtime governance for AI agents, with local policy decisions, pre-action
enforcement, and an authorization ledger.

Kontext runs alongside AI agents on developer machines and in cloud
environments. It receives tool-use events through local hooks, evaluates
policy before consequential actions execute, and records decisions and outcomes
in an authorization ledger. The decision path stays local; managed deployments
can export redacted records to the Kontext dashboard.

The agent support matrix is authoritative for each agent and event
surface. Blocking is off by default and available only at supported synchronous
pre-action hooks.

Quickstart

brew install kontext-security/tap/kontext
kontext setup

Create an install token in the Kontext dashboard when setup asks for one.
Setup stores the token in the login keychain, installs hooks for supported
agents, and starts a background daemon.

kontext doctor

Use doctor to check self-serve hook status, daemon health, and the managed
export backlog. It exits non-zero when a configured installation is unhealthy.
For a self-serve stale daemon, kontext doctor --fix performs the verified
restart; other findings include their manual remediation. Re-run kontext setup to rotate the token. Run kontext setup --uninstall to remove the
self-serve installation. Self-serve setup currently supports macOS.

How it works

agent asks to use a tool
        |
        v
Kontext receives the action through a hook
        |
        v
local policy allows, denies, or records a would-decision
        |
        v
the decision is written to the local authorization ledger

The decision path is local. A managed deployment adds configuration and record
export; it does not require a hosted service to answer every tool call.

Core features

Kontext balances security and utility for AI agents: low-risk actions keep moving, and unsafe actions can be blocked before they execute.

  • Observe agent actions. Record supported tool calls, policy decisions, and
    outcomes in a local authorization ledger.
  • Apply policy before actions run. Use deterministic rules for boundaries
    such as destructive commands, sensitive files, production systems, data
    exports, and credential access.
  • Roll out safely. Observe mode shows what policy would deny without
    interrupting work. Enforce mode returns a real denial for matching rules.
  • Keep evidence. Store redacted records locally and, for managed
    deployments, send them to the Kontext dashboard for review.
  • Run where the agent runs. Use Kontext on a developer machine, in a cloud
    sandbox, or in another managed agent environment with a supported hook.

Managed deployments

For enterprise identity, audit retention, organization controls, deployment planning, custom usage volume, and onboarding for security and platform teams, contact [email protected] or book here.

Agent support matrix

See the agent support matrix for the exact events, enforcement
points, installation scope, and known gaps for each agent. It is the source of
truth for what “supported” means; an integration is not treated as fully
covered merely because Kontext can receive an event from it.

Data handling

Kontext stores tool activity and decision evidence, not model reasoning or full
conversation history. Sensitive values are redacted before local storage and
managed export.

Development

go build -o bin/kontext ./cmd/kontext
go test ./...
go test -race ./...
go vet ./...

Community

Reviews (0)

No results found