Speakeasy – AI Control Plane for MCP, Agents and Skills

TL;DR

TL;DR: Speakeasy is an AI control plane that governs every MCP connection, skill and agent session in one place — giving enterprises connection management, identity, guardrails and observability across their entire AI stack.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: speakeasy.com ← MUST visit and verify
• Resource (AI control plane explainer): speakeasy.com/resources/ai-control-plane ← MUST read
• HN thread (TypeScript SDK generator): news.ycombinator.com/item?id=39191445
• License: Proprietary (SaaS); open-source SDK generator available at github.com/speakeasy-api
• Source last checked: 2026-07-16

What Is Speakeasy?

Speakeasy describes itself as “the AI control plane to connect, secure, and govern every agent.” Its core proposition is bridging the gap between AI policy written on paper and policy enforced at runtime.

The platform sits between your AI tools (Claude, GPT, Copilot, custom agents) and the external services they connect to — MCP servers, internal APIs, SaaS tools, databases. It provides four core pillars:

Connect — Unified management of MCP server connections and agent-to-service links. One place to configure, rotate and audit every outbound connection an AI tool can make.

Secure — Guardrails that block PII exfiltration, detect prompt injection attempts, and enforce data-loss prevention policies at the connection level. Every session is logged for compliance.

Control — Role-based access control (RBAC) and policy enforcement across every AI tool and agent in the organisation. Policy on paper becomes policy at runtime.

Observe — Session-level logging and observability dashboards. You can see which agents accessed which services, what data left, and when.

Speakeasy integrates with models from AWS, OpenAI, Anthropic, Azure, Oracle and others. It also has a product specifically for MCP gateway management — centralising MCP server discovery, authentication and traffic routing.

Setup Workflow

Prerequisites

• A Speakeasy account (free tier available)
• At least one MCP server or agent you want to govern
• Node.js 18+ if using the local SDK

Step 1: Create a Workspace

Sign up at app.speakeasy.com and create a workspace for your team or organisation.

Step 2: Connect Your First MCP Server

1. Navigate to Connections → New Connection
2. Select your MCP server type (custom, AWS, Azure, etc.)
3. Provide the server endpoint and authentication credentials
4. Speakeasy will validate the connection and surface what tools and resources are available

Step 3: Configure Policies

Define what data can leave your environment:

1. Go to Policies → New Policy
2. Set rules for specific agents or connection types
3. Enable guardrails such as PII blocking or prompt injection detection
4. Assign the policy to relevant workspaces or users

Step 4: Onboard Agents

Point your AI tools (Claude, OpenAI, etc.) to the Speakeasy MCP gateway instead of direct connections. The gateway proxies traffic, enforces policies, and logs everything — transparently to the end user.

Deeper Analysis

Who Is It For?

Speakeasy targets enterprise teams deploying AI at scale. If you are running multiple AI agents or giving employees access to AI tools that connect to external services, you need governance — not just at the model level but at the connection level. That is Speakeasy’s niche.

The open-source TypeScript SDK generator on the other side is for API teams that want to ship type-safe SDKs from an OpenAPI spec. That is a separate product, but both share the Speakeasy brand.

MCP Gateway

The MCP Gateway product is notable. Model Context Protocol (MCP) is becoming the standard for connecting AI agents to tools, but managing MCP server discovery and authentication at scale is still an unsolved problem for many teams. Speakeasy’s gateway handles this centrally — rather than configuring each agent individually.

Comparison to Alternatives

| Feature | Speakeasy | Native MCP | Manual RBAC | |---|---|---|---| | Centralised connection management | Yes | No | No | | Runtime guardrails | Yes | No | No | | Session observability | Yes | Limited | No | | MCP-native | Yes | Yes | No | | Enterprise pricing | Yes | N/A | Varies |

Practical Evaluation Checklist

• Does your organisation run more than one AI agent or tool?
• Do those agents connect to external services (APIs, databases, MCP servers)?
• Do you need to enforce data-loss prevention or PII blocking at the AI layer?
• Do you need audit logs for AI tool usage for compliance?
• Is your team using multiple AI providers simultaneously?

If you answered yes to two or more of the above, Speakeasy solves a real problem.

Security Notes

• All session logs are stored with Speakeasy — review their data retention and privacy policy before enabling full logging
• Guardrail effectiveness depends on correct policy configuration; test policies in staging first
• API keys for connected services are stored encrypted — rotate credentials regularly

FAQ

Q: Does Speakeasy work with self-hosted models? A: Speakeasy integrates with models via standard APIs (OpenAI-compatible). Self-hosted models that expose an API endpoint can be connected through the same gateway mechanism.

Q: Is this the same as the open-source TypeScript SDK generator? A: No. The open-source SDK generator at speakeasy-api is a separate product that converts OpenAPI specs into typed SDKs. Speakeasy the AI control plane is a SaaS product for AI governance.

Q: What is the pricing model? A: Speakeasy offers a free tier with limited connections. Enterprise plans are available with advanced guardrails and audit capabilities. Check speakeasy.com/pricing for current details.

Q: Does it work with Claude, GPT-4, Copilot and custom agents simultaneously? A: Yes. Speakeasy sits at the connection layer, so any AI tool that makes outbound connections through the MCP gateway can be governed uniformly.

Conclusion

Speakeasy fills a specific gap in the enterprise AI stack — the layer between AI agents and the services they call. As AI tools proliferate inside organisations, the need for connection-level governance, guardrails and observability grows accordingly. The MCP Gateway product alone is worth evaluating if your team manages multiple MCP servers.

If you are running a small personal AI setup, Speakeasy is probably overkill. But for teams where AI tool usage needs to be auditable, controllable and secure at runtime, it addresses a problem that is otherwise solved with a patchwork of ad-hoc policies.

Explore it at speakeasy.com.