The fast path to aworld-class MCP.

Speakeasy makes building and hosting an MCP server easy. Run MCP on our serverless infra. OAuth, identity, logging are all handled, and your server is compliant with the latest MCP spec.

Deploy for freeTalk to us

Trusted by

The biggest companies
Google
Fifth Third
Mistral
Cisco
The fastest startups
PlanetScale
Vercel
Polar
MoonPay

MCP basics,
and complexity handled for you

Speakeasy generates, hosts, and secures your MCP server. It also keeps up with enterprise auth, customer-level logs, and a spec that keeps changing, so your team stays on the product.

Deploy for free

Basics

What every production MCP server needs on day one.

Generate and customize

Every API operation becomes a tool, regenerated on each push from CI. Add TypeScript where a workflow needs custom logic.

Scale on serverless

Runs on the serverless platform that serves Google and PlanetScale, with no capacity to plan.

Stand up OAuth

Use your identity provider or Speakeasy as the authorization server. Works with every MCP client.

Complexity

What gets harder as adoption grows.

Deploy with Code mode

Agents write a short program against your tools and run it in a sandbox, so intermediate results stay out of the context window.

Connect your IdP

Every call is logged with the customer, tool, client, and result. Stream it to Datadog or any OpenTelemetry collector.

The latest spec, no migration

New protocol features ship on the platform, and your server picks them up without a rewrite.

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

PlanetScale

Make MCP a first-class artifact

See which tools agents call, from which clients, for which customers, and where calls fail. Every call also streams to your observability stack over OpenTelemetry.

Deploy for free
MCP & Tools

Monitor tool calls across every customer and client connected to your server.

Live

Tool calls

48,213

in the last 24 hours

Failures

289

0.6% of calls

Tools used

6

on 1 server

Customers

37

active in this window

Calls over time

Most used tools

list_projects15,428
get_deployment11,571
create_branch8,196
deploy_request6,268

Most used clients

ClaudeClaude19,767
ChatGPTChatGPT13,018
CursorCursor9,160
Claude CodeClaude Code6,268

Incoming calls

Claudelist_projects182ms
Cursorget_deployment96ms
ChatGPTcreate_branch341ms
Claude Codedeploy_requesterror
Claudelist_projects158ms

Customer stories

PlanetScale turned its platform API into a production MCP server and grew usage 44x in five months.

PlanetScale launched its MCP server on Speakeasy in February 2026. Five months later, agents were making more than 200,000 tool calls a month against it.

Read the PlanetScale story

44x

Growth in five months

From about 4,700 tool calls in the first month to 206,000 in the fifth.

200K+

Monthly tool calls

Developers route branch, schema, and deploy request work through their agents.

<24h

Issue turnaround

Every issue PlanetScale raised got a solution from Speakeasy in under a day.

Polar

Polar lets agents configure entire billing models through natural language.

Read the case studyFermat

Fermat launched a public MCP server so brands can query their Fermat data from any AI tool.

Read the case studyLaunchDarkly

LaunchDarkly extended runtime control to AI agents with an MCP server.

Read the case study

Explore more Speakeasy

Agent Identity

Every agent acts under a persistent identity.

Access governed by your identity providerShort-lived, task-scoped credentialsEvery decision attributed to a name
Explore Agent Identity

MCP Gateway

One governed path to every server.

One official registry of MCP serversShadow MCP discovered and promotedEvery tool call logged to your OTEL destination
Explore MCP Gateway

Agent Security

Every prompt and tool call checked.

Policy enforced on every actionPrompt injection, and data exfiltration protectionRunaway sessions quarantined
Explore Agent Security

Questions

What is a hosted MCP server?
An MCP server that Speakeasy generates and runs for you. Its tools come from your OpenAPI document, TypeScript functions, or both. Speakeasy implements the MCP protocol, executes each tool call against your API, and serves the server at a URL your users connect to, so there is nothing for your team to deploy, patch, or scale.
We already generate an MCP server from our API. What changes?
Speakeasy takes that generation over. Point it at the same OpenAPI document and it produces the tools on every push, then runs the server, puts OAuth in front of it, logs every call, and keeps it current with the spec. The generation script, the hosting, and the auth code you maintain today go away.
What does my team still own?
Your API and the business logic behind it, plus the decisions only you can make: which operations agents should see, how tools are named and described, and which workflows deserve a purpose-built tool written in TypeScript. Everything about running the MCP server is on Speakeasy.
Can the infrastructure handle our traffic?
Hosted servers run on Speakeasy's serverless infrastructure, the same platform that serves Google and PlanetScale. PlanetScale's server grew from about 4,700 tool calls in its first month to more than 200,000 in its fifth, on the same infrastructure.
Why ship a hosted server instead of a local one?
A local server asks every user to install a package and paste an API key into a config file, and it cannot run in web clients like Claude.ai or ChatGPT. A hosted server is a URL: users add it, sign in with OAuth, and every update reaches them without a reinstall.
How do my users authenticate?
Hook the server into the identity provider or authorization server you already run, such as Auth0 or your own, or let Speakeasy be the authorization server. The platform handles client registration for every MCP client. Enterprise Managed Authorization and verified clients are available when your customers ask for them, and servers can also accept API keys or bearer tokens directly. See secure with OAuth.
Can the server live on our own domain?
Yes. Verify a domain once and serve the server from an address on it, such as mcp.yourcompany.com. One address per domain can serve as the domain root.
How does the server stay in sync with our API?
Push your OpenAPI document from CI whenever the API changes. Each push produces a new immutable deployment, and the server serves the latest one. Removing an operation removes its tool, and a bad deployment rolls back in one step.
What happens when the MCP spec changes?
Speakeasy implements and maintains the protocol layer, so revisions such as the move to stateless MCP are handled on the platform, not in your codebase. Your team spends its time on the tools, not the transport.
Which clients can connect?
Claude, Claude Code, ChatGPT, Cursor, Copilot, Codex, and any other MCP-compatible client, along with the agents your customers build on agent frameworks. Every server has an installation page with setup steps for each client.
Can we also use hosted servers inside our company?
Yes. Hosted servers are part of the MCP gateway, so a private server for internal APIs gets IdP sign-in, team access, guardrails, and an audit log alongside the third-party MCP servers your employees use.
Can I run the server on my own infrastructure instead?
Yes. The Speakeasy CLI generates a standalone TypeScript MCP server from your OpenAPI document that you deploy and operate yourself. See standalone MCP server generation.

AI everywhere.

Control here.

Deploy for freeRead the docs