Getting started

Connect your AI agents to Guardian and govern every tool call they make.

Your workspace

Tenant: Loading
Tier: Loading
MCP endpoint: Loading
OAuth client ID: Loading

Connect an agent

Add Guardian to Claude Code

Run this in any project where your agent works. It registers Guardian as an MCP server, so the agent can call governance tools mid-task.

Other MCP clients work the same way: point them at your MCP endpoint over StreamableHTTP.

Authenticate with your OAuth credentials

Guardian requires OAuth 2.1 on every connection. Your client ID is shown above; your client secret was delivered once at provisioning and is never re-displayed here or anywhere in this portal. If the secret is lost, contact GuardianSupport@Pythagorithm.ai to rotate the credentials; rotation invalidates the old pair.

Verify the connection

Ask your agent to list its tools (in Claude Code: /mcp). A connected agent shows seven Guardian tools. Then make a harmless call, for example asking the agent to check a string with check_content_safety; the decision appears in Recent decisions within seconds, and the agent itself appears under Observed activity.

The seven governance tools

check_policyEvaluates an intended action against your tenant policy and returns allow, deny, or review with the reason.

check_content_safetyScreens content for prompt injection and unsafe patterns before your agent acts on it.

check_data_classificationClassifies data sensitivity so agents handle regulated or confidential material correctly.

check_agent_authorizationChecks whether a specific agent is sanctioned for a specific action against your allowlist.

request_approvalPauses for human sign-off; the request is delivered to your configured channel and the agent polls the verdict.

log_decisionRecords an agent's own decision into the tamper-evident audit chain.

explain_governance_actionExplains any recorded decision: the policy applied, the detector that fired, and the reasoning.

Granted scopes

Scopes are fixed per tenant at provisioning. Every tool call is recorded against your tenant's audit chain regardless of outcome.

Workspace setup hit a snag

Nothing is lost

You're on the list

What happens next

Access deactivated

What this means

Almost there

What happens next

Dashboard

No activity yet

Decisions over time

Outcome breakdown

Getting started

Your workspace

Connect an agent

Add Guardian to Claude Code

Authenticate with your OAuth credentials

Verify the connection

The seven governance tools

Granted scopes

Agents

Observed activity

Recent decisions

Admin

Controls

Leads

Unmanaged users

Approvals

Govern every move your AI agents make.

Sign in

Create your account

Confirm your email

Workspace setup hit a snag

Nothing is lost

One more step

You're on the list

What happens next

Access deactivated

What this means

Almost there

What happens next

Dashboard

No activity yet

Decisions over time

Outcome breakdown

Getting started

Your workspace

Connect an agent

Add Guardian to Claude Code

Authenticate with your OAuth credentials

Verify the connection

The seven governance tools

Granted scopes

Agents

Observed activity

Recent decisions

Admin

Controls

Leads

Unmanaged users

Approvals