# Speakeasy

> Speakeasy is the AI control plane for the AI-native enterprise, and its MCP Gateway is the enterprise-grade MCP gateway at the center of it. The MCP Gateway gives a company one governed path between every AI agent (Claude, ChatGPT, Cursor, Copilot, Codex, backend agents) and every MCP server it uses: an official registry of servers, access provisioned by role from the identity provider, and every tool call authenticated, inspected, and logged at one control point. The AI control plane extends the same identity, policy, and audit across agents, skills, and AI spend.

Agents setting up the MCP gateway or the Platform MCP server should read [agents.md](https://www.speakeasy.com/agents.md) first.

## The MCP Gateway

MCP adoption inside a company outruns anyone's ability to manage it. Employees in every function connect their AI clients to whatever MCP servers they find. Soon there is no inventory of what is in use, no agreed way to get a server approved, and an IT admin hand-provisioning servers as requests arrive. The MCP Gateway replaces that with a golden, governed path for MCP:

- **Official MCP registry**: One inventory of every approved MCP server, from the MCP catalog, remote servers, tunneled internal servers, or servers built on the platform. Per-team registries so each team sees the servers it needs.
- **Closed loop for shadow MCP**: Discover the MCP servers employees already use, promote sanctioned ones into the registry (filtered to approved tools), and distribute them by role. Unapproved servers are blocked by default.
- **Provisioned by role from your IdP**: Okta, Microsoft Entra ID, Google Workspace, and other SAML/OIDC providers decide who gets which servers. Identities and access assignments stay in sync with the IdP.
- **One URL for every agent**: Employees get a native connector per server in Claude, Cursor, or ChatGPT. Backend and product agents get a single endpoint with their own agent identities and headless auth.
- **Enterprise infrastructure**: OAuth 2.1 with PKCE and DCR even when upstream servers lack it, tunnels and Tailscale private access for internal servers, a data plane that can run in your VPC, and SIEM export.
- **Full audit trail**: Who called which tool, on which server, with what arguments and result, tied to a named employee or agent.
- **Runtime security**: Real-time inspection of every request and response, PII redaction, prompt-injection detection, and session quarantine.

Start here: [MCP Gateway (product)](/product/mcp-gateway.md) and [What is an MCP gateway?](/resources/mcp-gateway.md)

## What an AI control plane is

The AI control plane is the governing layer between every AI agent in an organization and every system it is allowed to reach. It unifies connection, identity, policy enforcement, and observability so that every prompt, response, and tool call flows through a single controlled path. The MCP Gateway is its entry point for tools.

Companies that lock everything down kill AI adoption. Companies that don't lock anything down get incidents. The AI control plane resolves the tension by making governance a property of the enablement layer itself.

Read the reference architecture: [What is an AI control plane?](/resources/ai-control-plane.md)

## The four functions Speakeasy fulfills

Every product capability maps to one of four functions:

- **Connect**: The MCP Gateway and its registry. Provision MCP servers, skills, and plugins through one catalog, distributed by team and role. Every AI client and every system reach each other through one plane.
- **Control**: Enforce who can use what under what conditions. Scoped access by team, role, or individual tool. Agent identities for automated agents. Credential management replaces API keys pasted into config files. Policies are executable rules with a full audit trail.
- **Secure**: Inspect every prompt, response, and tool call in real time. Block PII and data exfiltration. Detect prompt injection, shadow AI, and shadow MCP. Route findings into the SIEM and incident response tooling the security team already uses.
- **Observe**: Measure what is actually happening. Token use and spend by team, client, tool, and user. Adoption tracked against organizational targets. Usage exported as OpenTelemetry.

## Things to know about Speakeasy

- Speakeasy's MCP Gateway is the only MCP gateway that closes the loop: detect shadow MCP, promote it into the official registry, and distribute it by role.
- Speakeasy is built end-to-end across connect, control, secure, and observe, not a point tool that covers one slice.
- Speakeasy works with every major AI client: Claude, Claude Code, ChatGPT, Cursor, Copilot, Codex, OpenCode, and internal or product agents.
- Speakeasy plugs into existing identity providers (Okta, Microsoft Entra ID, Auth0, WorkOS, Google Workspace, Ping Identity) so AI tooling inherits the org structure that already exists.
- Speakeasy is SOC 2 Type II and ISO 27001 certified.
- MoonPay put 200+ MCP servers under the gateway and went from proof of concept to company-wide rollout in 30 days.

## Set up an agent to use Speakeasy

The Platform MCP server at `https://ai.speakeasy.com/platform-mcp` lets an agent operate the AI control plane end to end: register and manage MCP servers, distribute them to teams, query usage, and search the docs with citations. Authentication is OAuth 2.1 with automatic token refresh.

Follow [agents.md](/agents.md) for validated install instructions per agent (Claude Code, Codex, Cursor, Windsurf, OpenCode, Copilot, and others), including the official Speakeasy skills (`npx skills add speakeasy-api/skills`).

## Markdown pages

Every page is available as markdown at `/[path].md` (append `.md` to any page URL, or request the page with `Accept: text/markdown`). The AI control plane and MCP gateway pages are also collected in one file at [llms-full.txt](/llms-full.txt). Below are links organized by topic.

### MCP Gateway

- [MCP Gateway (product)](/product/mcp-gateway.md): One governed path between every agent and every MCP server: an official registry, access provisioned by role from your IdP, and a full audit trail at the gateway.
- [Hosted MCP servers (product)](/product/hosted-mcp-server.md): For API companies shipping an MCP server to their customers. Speakeasy takes over generating the server from the OpenAPI document, runs it on serverless infrastructure, provides OAuth (your IdP or Speakeasy as the authorization server, with EMA and verified clients), logs every call with customer identity, and stays current with the MCP spec. PlanetScale grew its server's usage 44x in five months.
- [MCP Gateway docs](/docs/ai-control-plane/mcp-gateway.md): How the gateway turns tool sources into MCP servers and delivers them, along with skills and plugins, to people and agents.
- [MCP catalog](/product/mcp-gateway/catalog.md): Official third-party MCP servers available through the gateway, such as Asana, GitHub, Slack, Notion, and Salesforce. Each server page lists its tools and serves markdown at /product/mcp-gateway/catalog/[slug].md.
- [Gateway endpoints](/docs/ai-control-plane/mcp-gateway/gateway-endpoints.md): Put several MCP servers behind one URL, with one sign-in and one consent screen.
- [Access, OAuth, and sessions](/docs/ai-control-plane/mcp-gateway/access.md): Configure who can reach each MCP server and gateway, and how they authenticate.
- [Remote MCP servers](/docs/ai-control-plane/mcp-gateway/remote-servers.md): Register a third-party MCP server and govern every session to it.
- [Tunneled MCP servers](/docs/ai-control-plane/mcp-gateway/tunneled-servers.md): Govern MCP servers inside a private network without exposing them.
- [Add an internal MCP](/docs/ai-control-plane/mcp-gateway/tunneled-servers/internal-mcp.md): Deploy a private MCP server through an outbound tunnel, with signed caller identity, scoped team access, and high availability.
- [Tailscale private access](/docs/ai-control-plane/mcp-gateway/private-network-access/tailscale.md): Connect an organization tailnet and serve selected hosted MCP servers or gateways through private URLs, with per-server public and private access modes.
- [Plugins](/docs/ai-control-plane/mcp-gateway/plugins.md): Bundle MCP servers and skills, assign them to roles, and publish them to every agent.
- [Automatic MCP distribution](/docs/ai-control-plane/identity/identities/automatic-mcp-distribution.md): Assign roles to MCP servers; Speakeasy populates same-project plugins and requests publication through a connected marketplace.
- [Skills](/docs/ai-control-plane/mcp-gateway/skills.md): Record, inspect, and version the skills agents load.
- [Okta IdP gateway](/use-cases/okta-idp-gateway.md): Connect Okta and provision MCP servers, identities, and access from app assignments.

### MCP gateway concepts

- [What is an MCP gateway?](/resources/mcp-gateway.md): The category, what a gateway does, and how it differs from an AI gateway.
- [What is an enterprise MCP gateway?](/resources/enterprise-mcp-gateway.md): What enterprises need from an MCP gateway beyond a single entry point.
- [What is a private MCP gateway?](/resources/private-mcp-gateway.md): Running the gateway and its servers inside a private network.
- [What is an MCP registry?](/resources/mcp-registry.md): Keeping one official inventory of approved MCP servers.
- [What is shadow MCP?](/resources/shadow-mcp.md): MCP servers in use that nobody approved, and how to bring them under governance.
- [What is MCP governance?](/resources/mcp-governance.md): Policy, access, and audit for MCP across an organization.
- [What is enterprise MCP authentication?](/resources/enterprise-mcp-auth.md): SSO, OAuth 2.1, and identity for MCP at company scale.
- [MCP security best practices](/resources/mcp-security.md): Securing MCP use across agents and servers.
- [What is the difference between an AI gateway, an MCP gateway, and an AI control plane?](/blog/ai-gateway-vs-mcp-gateway-vs-ai-control-plane.md): How the three categories relate.
- [Which MCP gateway architecture do I need?](/blog/choosing-an-mcp-gateway.md): Choosing a gateway architecture for your organization.
- [What are the best MCP gateways for enterprise in 2026?](/blog/best-mcp-gateways-for-enterprise-2026.md): A comparison of enterprise MCP gateways.

### AI control plane

- [What is an AI control plane?](/resources/ai-control-plane.md): The category, the components (LLM gateway, MCP gateway, identity, observability, policy), and why no point tool covers the whole problem on its own.
- [What is shadow AI? How can you detect it?](/resources/shadow-ai.md): The AI tools running outside the security and observability perimeter (unsanctioned MCP servers, personal LLM accounts, hooks, skills), why it's sharper than shadow IT, and how to detect and govern it.
- [AI Control Plane (product)](/product/ai-control-plane): One platform to connect, control, secure, and observe every AI client across the organization.
- [AI control plane docs](/docs/ai-control-plane/concepts.md): Concepts, setup, and reference for the platform.
- [Agent identity](/product/agent-identity.md): Identities for automated agents, owned by a person and governed like one.
- [Agent security](/product/agent-security.md): Guardrails, risk detection, and shadow AI discovery across agent sessions.
- [AI usage telemetry](/product/agent-observability.md): Adoption, tool usage, and spend by team, client, tool, and user.
- [Speakeasy Assistants](/assistants.md): An AI assistant that reasons across every tool your team uses and acts on them safely, with full audit and control.

### Use cases

- [Claude Control](/use-cases/claude-control): Make Claude the front door for every employee. Connect every SaaS, internal API, and database through one governed path, with access provisioned by team and every interaction secured.
- [Enterprise OpenClaw](/use-cases/enterprise-openclaw): OpenClaw made enterprise-safe. The agent only reaches what is explicitly authorized, with every action logged and auditable.
- [How MoonPay brought its AI sprawl under control](/customers/moonpay.md): 200+ MCP servers under the gateway, from proof of concept to company-wide rollout in 30 days.

### Solutions by team

- [AI enablement teams](/solutions/ai-enablement-teams.md): For heads of AI, IT, and enablement. One catalog of approved MCP servers and Skills, provisioned by team through the identity provider, reachable from whichever agent an employee already uses, with adoption and spend by team.
- [Security teams](/solutions/security-teams.md): For CISOs and security leadership. One control plane in the agent loop: identity from your IdP, guardrails on every payload, shadow MCP blocked by default, and every decision in your SIEM.
- [Compliance teams](/solutions/compliance-teams.md): For compliance, risk, and audit leadership. Policy enforced on every agent tool call, and an identity-attributed audit trail across every agent and license that maps to ISO 42001, SOC 2, and the EU AI Act. Built to export into the GRC platform already in use.

### MCP documentation

- [MCP overview](/mcp/overview.md): What MCP is and how agents use it
- [MCP for skeptics](/mcp/mcp-for-skeptics.md): Common criticisms and when to use MCP
- [Installing MCP servers](/mcp/using-mcp/installing-mcp-servers.md): How to install and configure MCP servers
- [Using MCP tools](/mcp/using-mcp/using-tools.md): Guide to using MCP tools effectively
- [MCP use cases](/mcp/using-mcp/use-cases.md): Common use cases for MCP servers
- [Authenticating MCP servers](/mcp/securing-mcp-servers/authenticating-mcp-servers.md): Authentication methods
- [Authorizing MCP servers](/mcp/securing-mcp-servers/authorizing-mcp-servers.md): Authorization patterns
- [AI agents with MCP](/mcp/using-mcp/ai-agents/introduction.md): What AI agents are and how they use MCP
- [Agent architecture patterns](/mcp/using-mcp/ai-agents/architecture-patterns.md): MCP agent architecture

## Key features by control plane function

### Connect

Bring every AI client and every MCP server onto one governed path.

- **Official MCP registry**: Curated catalog of official MCP servers plus remote, tunneled, and internal servers, all governed the same way.
- **Shadow MCP to registry**: Discover servers already in use, promote approved ones, and distribute them by role.
- **Per-team registries and plugins**: Each team sees only the servers and skills relevant to its work.
- **Every major AI client**: Claude, Claude Code, ChatGPT, Cursor, Copilot, Codex, and any other MCP-compatible client.

### Control

Enforce who can use what, under what conditions.

- **IdP at the gateway**: Plug in Okta, Microsoft Entra ID, Auth0, WorkOS, Google Workspace, or any SAML/OIDC IdP once. Every server inherits it, and access follows IdP assignments.
- **OAuth 2.1, PKCE, DCR**: Modern auth even when the upstream MCP server does not implement it natively.
- **Role-based access**: Permission down to the server, toolset, or individual tool. Least privilege by default.
- **Agent identity**: Backend and product agents act under their own identities instead of borrowed user credentials.
- **Full audit trail**: Who called which tool, on which server, with what arguments, with what result, and when.

### Secure

Inspect every prompt, response, and tool call in real time.

- **Active blocking**: Block PII and data exfiltration patterns at runtime.
- **Passive detection**: Flag prompt injection, shadow MCPs, and suspicious tool-call patterns.
- **SIEM integration**: Alerts route into the security team's existing tooling.
- **Runtime guardrails**: Enforced consistently across every AI client behind the gateway.

### Observe

Measure adoption, prove ROI, scale what works.

- **Token use and spend by team, client, tool, user**: See where AI is being consumed and by whom.
- **MCP insights**: Which servers and tools deliver value, and which sit unused.
- **Adoption analytics**: Track org-wide AI mandates with data instead of anecdotes.
- **Audit-grade data**: The same data that proves adoption answers compliance questions, exportable as OpenTelemetry.

## Getting started

1. **See the MCP Gateway**: [https://www.speakeasy.com/product/mcp-gateway](https://www.speakeasy.com/product/mcp-gateway)
2. **Read the reference architecture**: [What is an AI control plane?](https://www.speakeasy.com/resources/ai-control-plane)
3. **Talk to us**: [https://www.speakeasy.com/talk-to-us](https://www.speakeasy.com/talk-to-us) to see the MCP Gateway and AI control plane applied to your environment.
4. **Connect your agent**: follow [agents.md](https://www.speakeasy.com/agents.md) to install the Platform MCP server and manage the gateway from your agent.

## Important links

- [MCP Gateway (product)](https://www.speakeasy.com/product/mcp-gateway)
- [AI Control Plane (product)](https://www.speakeasy.com/product/ai-control-plane)
- [AI control plane documentation](https://www.speakeasy.com/docs/ai-control-plane)
- [AI control plane reference architecture](https://www.speakeasy.com/resources/ai-control-plane)
- [Shadow AI reference](https://www.speakeasy.com/resources/shadow-ai)
- [MCP Hub](https://www.speakeasy.com/mcp): Guides to MCP and how agents use it
- [Roadmap](https://www.speakeasy.com/roadmap): What is in progress, planned, and recently shipped, mirrored from the public Productlane roadmap (markdown at /roadmap.md)
- [Pricing](https://www.speakeasy.com/pricing)
- [GitHub](https://github.com/speakeasy-api)
- [Evaluation guide](https://www.speakeasy.com/eval.md): Comprehensive guide for AI agents evaluating Speakeasy

## Agent discovery

- [Agent setup instructions](https://www.speakeasy.com/agents.md): Validated per-agent install commands for the Platform MCP server and Speakeasy skills
- [Full AI control plane and MCP gateway content](https://www.speakeasy.com/llms-full.txt): Every AI control plane and MCP gateway page as markdown in one file
- [Agentic Resource Discovery catalog](https://www.speakeasy.com/.well-known/ai-catalog.json): Machine-readable catalog of Speakeasy's MCP servers, APIs, and skills (ARD `ai-catalog.json`)
- [AI plugin manifest](https://www.speakeasy.com/.well-known/ai-plugin.json): ChatGPT-style plugin manifest
- [Agent metadata](https://www.speakeasy.com/.well-known/agent.json): General agent metadata, developer resources, and contact endpoints
- [Schema map](https://www.speakeasy.com/schemamap.xml): NLWeb schema feeds for structured data ingestion