How does Speakeasy handle shadow AI?

It maintains one catalog of approved tools. Connections outside this catalog are detected, quarantined, and flagged, requiring formal approval to be brought into the governed path.

How are policies enforced?

Every prompt, response, and tool call routes through the control plane, where it's checked against policy in real time before reaching your systems. All allow/deny decisions are logged.

Does Speakeasy replace my identity provider?

No. It extends your existing IdP (Okta, Entra ID, Auth0, etc.) to give AI agents their own identity, scoping access to your existing teams and roles.

What is the 'Agent-friendly docs' feature?

The documentation is optimized for AI consumption, with raw markdown available at every URL, an llms.txt index, and a full llms-full.txt dump for agents to read directly.

How do custom tools impact request-based billing?

A single custom tool call that orchestrates multiple sub-tool calls counts each sub-call as a separate request toward your usage quota.

Can I self-host Speakeasy?

Yes. A self-hosted option is available for deployment in your own cloud or on-premises for full data residency control with no external dependencies.

Directory/Automation/AI Governance/Speakeasy

Speakeasy

Speakeasy is an AI control plane designed for enterprises to securely scale AI usage. It provides a single layer to understand how AI is used across the company, define what agents and tools can access, and enforce security policies on every action. Key capabilities include a unified catalog for approved AI tools, identity-aware access control that extends your existing IdP to agents, and real-time policy enforcement that blocks threats like prompt injection and data leaks. The platform helps security, compliance, and AI enablement teams eliminate shadow AI, enforce least-privilege access, and maintain a full audit trail for governance and compliance. It integrates with major identity providers and deploys via MDM for fleet-wide coverage.

Automation

Visit website

Added on

Sep 21, 2026

AI Control PlaneAgent GovernanceMCP SecuritySSO IntegrationAudit LogsShadow AI DetectionPolicy Enforcement

Product information

Everything worth knowing about Speakeasy

The complete picture — from everyday features to the technical detail builders and IT folks dig for.

Core features

What it actually does

Centralized AI Governance

Single control plane to understand, define, and enforce policies for all AI agents, MCP servers, and skills across the company.

Agent Identity & Access Control

Extends your existing identity provider (Okta, Entra ID, Auth0, etc.) to AI agents, granting them unique identities and role-based access.

Real-time Security Enforcement

Checks every prompt, response, and tool call against policy before it reaches your systems, blocking threats like prompt injection and PII exposure.

Full Audit Trail

Logs every allow-or-deny decision for compliance and forensic analysis.

Shadow AI Detection

Maintains a catalog of approved tools and detects/quarantines connections to unapproved MCP servers.

Technical capabilities

MDM Deployment

Deploys fleet-wide through existing MDM solutions (Jamf, Intune) for agent coverage from day one.

MCP Protocol Support

Governs any agent (Claude, ChatGPT, Cursor, etc.) that supports the Model Context Protocol (MCP).

Must watch videos

Launch quality MCP servers in seconds | Speakeasy AI Demo at infra.nyc (June 26)

8:58

Tutorial8:58

Launch quality MCP servers in seconds | Speakeasy AI Demo at infra.nyc (June 26)

✨ Brian Flad - Founding Engineer, Speakeasy sharing their newest and most exciting product for MCP creation, curation and hosting, at infra.nyc! 𝗚𝗿𝗮𝗺 is new platform that helps companies to build and launch high quality and official MCP servers. Speakeasy.com is a full API development platform that transforms your OpenAPI specs into polished SDKs, Terraform providers, developer-friendly docs, and MCP (agent-ready) servers—built for teams and the AI era.

0:59

Tutorial0:59

Speakeasy AI Control Plane

Introducing Speakeasy's AI Control Plane

Who uses it

Real use cases, no hype

Shadow AI Discovery & Quarantine

Detect unapproved agents and MCP servers, isolate them, and funnel approval requests through a governed path.

Enterprise Agent Governance

Define granular, role-based access policies for every AI agent action from a single control plane.

MCP Server Lifecycle Management

Create, deploy, version, and secure Model Context Protocol servers with built-in OAuth and RBAC.

AI Cost & Usage Observability

Track agent sessions, tool calls, token usage, and spending in real time across the organization.

Compliance & Audit Readiness

Generate an immutable audit trail for every allow/deny decision to satisfy SOC 2, ISO 27001, and EU AI Act requirements.

Secure AI Rollout via MDM

Deploy agent controls and security policies fleet-wide from day one using Intune, Jamf, or other MDM.

Centralized MCP & Skill Catalog

Maintain a single, governed catalog of approved tools and skills for all AI agents across the company.

What's great

Centralized governance for all AI agents and MCPs across the enterprise.
Seamless integration with existing identity providers (Okta, Entra ID, etc.)
Real-time policy enforcement on every agent action, prompt, and tool call.
Comprehensive audit trail and logging for compliance (SOC 2, ISO 27001).

Technical strengths

Deploys fleet-wide via existing MDM (Jamf, Intune).
Extends enterprise IAM to agent identity with RBAC based on existing teams/roles.
Blocks prompt injection, PII exposure, and leaked credentials at the control plane.

Where it falls short

Primarily focused on large enterprises; may be overkill for small teams.
Enterprise-only pricing with no transparent public tier; requires sales contact.
Adds a mandatory proxy layer between all agents and systems.

Technical limitations

Depends on MDM and SSO providers for full deployment and policy inheritance.
Custom tool call orchestration can inflate request-based billing.
Agent support is still expanding (e.g., Mistral, Gemini listed as 'Coming soon').

For technical folks