AI & MCP

Announcing the Speakeasy MCP Gateway: one entry point between every AI client and every MCP server

Sagar Batchu

March 10, 2026 - 5 min read

---

From Gram to the Speakeasy MCP Gateway

We started Gram as the open source platform  for building MCP servers. The thesis was straightforward: to power AI applications, teams needed a fast path to a deployed MCP server, and we could provide it.

We were right about the importance of MCP servers, but building them turned out not to be the hard problem. After months of design partner work and dozens of customer conversations, we learned that the real blockers to MCP in production were auth, observability, and distribution (deep dive on our learnings).

Today we’re launching the Speakeasy MCP Gateway: one entry point between every AI client and every MCP server, with SSO, policy, observability, and audit at the gateway.

Your MCP control plane

We believe that every employee should be using Claude, or their preferred AI app, as their single interface for all work. No more bouncing between dashboards and SaaS apps.

We’ve built the Speakeasy MCP Gateway to make that transformation possible. It’s the control plane that connects every employee in your organization to every MCP server with the security, governance, and observability you need to do it safely at scale.

Here’s what that looks like.

Build — generate, import, deploy, distribute

Most teams have dozens of APIs and SaaS tools that need to be accessible to AI. Building MCP servers for each one by hand doesn’t scale. The Speakeasy MCP Gateway gives you multiple paths to a production server: generate from an API spec, import an existing server, or deploy a pre-built integration. However you build, we’ll handle deployment and distribution from there.

Build servers fast. Use an existing API, import an existing TypeScript server to get gateway capabilities without rewriting anything, or deploy a pre-built integration like Stripe, GitHub, or Slack from the SaaS catalog in minutes.

Dynamic servers for token efficiency. Servers behind the Speakeasy MCP Gateway support progressive disclosure, by surfacing two tools: search_tools, execute_tool. Same functionality, better token efficiency.

One-click rollout across your org. The gap between “server exists” and “people use it” has been one of the biggest unspoken problems in the MCP ecosystem. The Speakeasy MCP Gateway closes it: deploy a server and make it available to your entire organization in a single step. Servers can be provisioned as Claude plugins automatically, and they work with Claude Web, Claude Desktop, Cursor, Windsurf, and any other MCP-compatible client.

“Speakeasy made it trivial to turn our database platform API into a production MCP server. Developers interact with branches, schemas, and deploy requests through AI agents instead of dashboards.”

— Mike Coutermarsh, PlanetScale

Secure — unified auth for the whole team

Security was the number one blocker we heard from every customer segment. Teams were building MCP servers and watching them sit unapproved for months while auth got sorted out. The Speakeasy MCP Gateway makes auth a platform feature, not a configuration exercise. Connect your SSO to our OAuth proxy, scope access per team, and skip the months of auth plumbing.

Unified auth layer. Every MCP server in your org gets the same auth experience regardless of how the underlying service connects. OAuth, API keys, custom tokens, Speakeasy abstracts it all behind a single OAuth 2.1 layer with DCR and PKCE. Plug in your SSO once, and every server is covered.

Scoped access. Role-based permissions at the server, toolset, and individual tool level. Provision sub-catalogs so every team sees exactly what they need.

Client hooks. Get visibility across your entire org’s AI usage by intercepting MCP traffic at the client level. Identify shadow servers, enforce policies, and ensure every MCP connection flows through your gateway.

“I was able to equip the entire solutions engineering org with MCP servers in a single afternoon. Speakeasy handled auth and permissions, so we didn’t have to choose between speed and security.”

— Eli Davis, Fivetran

Observe — see what’s happening in production

Teams told us they were flying blind once servers hit production. They could see that tool calls were happening. They couldn’t see why calls were failing, which server version handled a request, or whether agents were using tools the way they’d intended.

Real-time telemetry and distributed tracing. Live visibility into every tool call, every session, every server across your organization. Trace requests across servers and sessions to understand the full path of an agent interaction, not just individual tool calls in isolation.

Performance metrics per tool. Latency, error rates, and usage patterns broken down by individual tool. Understand adoption across your organization and identify bottlenecks before they become incidents.

Actionable failure insights. Not just dashboards that tell you something broke, but context that helps you understand why and fix it. Error categorization, version correlation, and root cause signals.

“We needed to understand how AI was being adopted across the company and where people were getting stuck. Speakeasy gave us that visibility from day one.”

— Shreyas Kumar, Co-founder, Fermat

Get started

The Speakeasy MCP Gateway is live today. If you’re rolling MCP out across your org, or trying to run servers in production, this is the gateway that makes it work.

Get started at speakeasy.com/mcp 

For the full backstory on how we got here, read We were wrong about the hard problem.

The Speakeasy MCP Gateway is the control plane between every AI client and every MCP server, with SSO, policy, observability, and audit at the gateway. speakeasy.com/mcp