Docs
DocumentationOverview

Kastra documentation

Understand Kastra’s security boundary, prove a workflow, and run it with clear ownership.

Guides, integrations and API reference Read as Markdown
On this page

Start with the decision your team needs to control. Security leaders can review architecture and assurance; engineers can connect a supported integration; operators can plan rollout and response.

Choose your path

Review Kastra’s SOC 2 assurance and GDPR documentation in the Trust Center, alongside the product behavior described in these guides.

Open the Kastra Trust Center ↗

Connect an integration

Coding agents

Install Edge and connect a supported coding-agent hook.

Get started →Applications

Govern supported OpenAI or Anthropic traffic through the proxy.

Get started →MCP tools

Wrap a local stdio server and test its tool permissions.

Get started →

Start here6 guides

How Kastra worksChoose the integration that can see and govern the action you care about.Core concepts and decision statesUnderstand the vocabulary used in setup, policy, approval, and evidence guides.Your first governed workflowConnect, observe, test a block, and resolve an approval in a test environment.Authentication and environmentsUse the credential and environment contract for the surface you are connecting.Platform and integration supportDistinguish available artifacts from combinations validated on real machines.Find your way around the consoleNavigate the current application from setup to policy, review, evidence, and administration.

Security review7 guides

Evaluate Kastra for your security programA CISO's guide to the controls, evidence, and operating decisions needed to evaluate Kastra.Architecture and enforcement coverageTrace each action through its enforcement point, policy service, and evidence path.Data protection and privacy reviewMap the data each integration processes and the privacy evidence needed before a rollout.Access control and shared responsibilitiesAssign policy owners, approvers, and credential responsibilities before connecting sensitive workflows.Pilot plan and acceptance criteriaRun a bounded evaluation with observable control outcomes and a clear rollout decision.Rollout, monitoring, and incident responseOperate the integration, respond to non-enforcement, and preserve a practical recovery and exit path.SOC 2, GDPR, and the Trust CenterFind Kastra's vendor assurance material and connect it to your security and privacy review.

Coding agents9 guides

Install Kastra EdgeInstall the correct platform build, sign in, and connect your agent hook.Claude CodeConnect Claude Code to Kastra policies and verify the hook boundary.Codex CLIConnect Codex CLI to Kastra policies and verify the hook boundary.CursorConnect Cursor to Kastra policies and verify the hook boundary.OpenClawInstall the Kastra plugin and govern supported OpenClaw tool calls.Hermes AgentConnect native and MCP tool hooks, including Hermes first-use consent.Desktop app and device managementUse desktop approvals and manage devices without confusing the daemon with hook enforcement.Usage telemetry and collectionUnderstand Claude Code and Codex telemetry installed alongside hooks and its reporting limits.Troubleshoot EdgeFind missing hooks, wrong binaries, environment mismatches, and fail-open events.

Applications & proxy6 guides

Proxy setupRoute supported OpenAI or Anthropic traffic through Kastra with both credentials.OpenAI Chat CompletionsSend a server-side Chat Completions request with explicit Kastra and upstream credentials.Anthropic MessagesConnect the Messages protocol through Kastra and handle policy outcomes.Streaming and model outputHandle filtered output, tool-call decisions, and failures after a stream has started.Proxy HOLD and resubmissionUse the correct approval contract for held requests and generated tool calls.Content scanning and redactionApply supported secret and PII detectors to proxy traffic with visible scope.

MCP7 guides

Choose the right MCP connectionSeparate downstream tool governance, account inspection, and onboarding.Local MCP gatewayWrap an existing stdio server and verify discovery, deny, and approval behavior.Tool catalogs and permissionsUse real tool names in policies and enforce restrictions at discovery and call time.Account MCP connectorRead governance state through OAuth or the local read-only server.Onboarding MCPLet an assistant start a signup handoff while the user completes account creation.Claude Code pluginInstall read-only governance skills and follow the separate Edge enforcement setup.Hosted MCP availabilityUnderstand the separate release boundary for remote third-party MCP connections.

Policies6 guides

Author and test policiesCreate a rule that matches the real integration surface and activate it deliberately.Policy as codeValidate, apply, version, and promote .kastra policies with the operator CLI.Safety packsStart from curated rules while preserving surface scope and exemptions.AI policy draftingDescribe a rule, review catalog matches and tests, and install the result yourself.Recon history scanningTurn supported Claude Code and Cursor history into reviewable policy recommendations.Shadow mode and replayEvaluate policy impact before enforcement and understand historical coverage.

Approvals5 guides

Approval lifecycleUnderstand pending checkpoints, human resolution, expiry, and effective decisions.Smart HoldsFlag false positives and review proposed policy refinements without automatic changes.Slack approvalsConnect a workspace, choose notification channels, and link each human approver.Signed webhooks and notificationsReceive governance events, verify authenticity, and handle retries.Notifications and alert rulesSet personal notification preferences and administrator alert rules.

Evidence & usage6 guides

Decision history and verificationInspect the exact action, policy, and outcome, then verify the recorded chain.Framework evidence exportsGenerate scoped reports, verify their seals, and give reviewers a clear account of coverage.Data handling and maskingSeparate evaluation payloads, stored evidence, Recon uploads, and usage reporting.Usage and cost provenanceRead metered and reported usage without treating missing data as zero or estimates as invoices.Rate limits and spend capsApply windowed controls on supported paths and understand identity and concurrency limits.Governance audit logInspect administrative changes separately from runtime decisions and notification delivery.

Administration7 guides

Workspaces, roles, and environmentsKeep membership, policy scope, and environment identity explicit.Agent inventory and identityDefine matchable selectors and tool scope without implying attested workload identity.SAML SSO and SCIMConfigure Enterprise identity with explicit protocol and provisioning boundaries.Plans and observe-only behaviorUnderstand the Pro trial, feature entitlements, and what changes when billing lapses.API keys and identity bindingCreate a scoped runtime credential and manage its binding and revocation.Devices and operator tokensReview workspace connections and revoke credentials you no longer use.Account security and MFAEnroll an authenticator and distinguish account security from workspace identity settings.

Reference7 guides

Operator CLI referenceManage policies, evidence, environments, keys, and checkpoints from a terminal or CI.Evaluation APISubmit an action, decode ALLOW or DENY, and negotiate a supported HOLD response.Policy attributes and effectsMatch the context the integration actually sends, with explicit surface and trigger scope.Failure behaviorDistinguish policy denial, unavailable evaluation, billing observe-only, and host timeouts.TypeScript SDK statusUnderstand the alpha implementation and its cooperative enforcement boundary.Self-hosted deployment and availabilityImplemented appliance capabilities, offline licenses, access recovery, and the requirements for a customer-operated deployment.API reference and contractsFind the reviewed runtime endpoints and navigate the full implementation schema.

The TypeScript SDK is in alpha; public package installation is not verified. Use the evaluation API for a custom integration and review the SDK status before planning adoption.

SDK status and integration boundary