Teams / AI enablement

Build a golden pathfor every agent.

The Speakeasy MCP gateway gives the company one governed path: a catalog of approved MCP servers and Skills behind one URL, provisioned by team through your identity provider, and reachable from whichever agent an employee already uses.

Explore the platformTalk to us

01 / The problem

Without a golden path,
AI programs fail

Demand for AI arrives before the path for it exists. Employees wire up what they can, security blocks what it cannot see, and the usage that matters most ends up on accounts the company does not own.

Shadow AI

Connected without oversight

Employees connect agents, MCP servers, and Skills to company data on their own: agents on laptops, servers pulled from the internet, Skills shared in chat, with no review and no owner. Nobody can list what is connected, so nothing on the list can be approved.

Blocked

Security is blind and defensive

Without identity or an audit trail, every agent request looks like a shared account reaching production. The safe answer is no, and the rollout waits behind it.

Personal licenses

Usage shifts to personal plans

Personal Claude and ChatGPT plans and free-tier coding tools have no admin console, no SSO, and no audit export. The employee gets the model they want, and the organization gets no visibility into what it touches.

02 / The solution

You need a control plane.

One governed entry point between every agent and every system it touches. Your platform team curates a catalog of approved MCP servers behind it, access follows your identity provider, and approving a tool once makes it available to everyone who should have it.

Team by team

Every tool is a project

  • Each team wires its own MCP servers with its own keys
  • Each server waits on its own security review
  • Nobody can list what agents are connected to
  • Adoption is an anecdote when the board asks for a number

Speakeasy MCP gateway

Approve once, distribute everywhere

  • One gateway URL for every agent, with a catalog your platform team curates
  • A server approved once inherits auth, policy, and logging for every team
  • Access provisioned by team and role from your identity provider
  • Enrollment, sessions, and spend by team, so the next rollout runs on numbers

03 / The catalog

A catalog of approved MCP servers

Browse and connect to MCP servers through the MCP gateway. Every server in the catalog is vetted and inherits the gateway's auth, RBAC, and audit logs. Bring your own servers and they get the same controls.

GitHubOfficialConnect AI assistants to GitHub - manage repos, issues, PRs, and workflows through natural language.Developer toolsSlackOfficialSearch messages and files, send messages, manage canvases, and access user profiles.Communication13 toolsNotionOfficialOfficial Notion MCP serverProductivity14 toolsLinearOfficialMCP server for Linear project management and issue trackingProject management31 toolsSalesforce PlatformOfficialAccess Salesforce CRM data and invoke platform actions via hosted endpoints.Sales & CRMHubSpotOfficialAccess and interact with HubSpot CRM data including contacts, companies, deals, and tickets.Sales & CRM13 toolsSnowflakeProvides Snowflake database access through a hosted MCP server built by Speakeasy and delivered via Gram's MCP hosting platform.Data & analytics3 toolsBigQueryOfficialGoogle-managed MCP server for BigQuery with SQL execution and dataset/table discovery.Data & analytics6 toolsFigmaOfficialThe Figma MCP server brings Figma design context directly into your AI workflow.Design17 toolsAsanaOfficialAccess Asana tasks, projects, goals, and team data from AI assistants.Project management44 toolsPagerDutyOfficialPagerDuty's official MCP server which provides tools to interact with your PagerDuty account.Infrastructure64 toolsAtlassian RovoOfficialEnables secure, permission-aware access to Atlassian Cloud products.Project management31 tools

45 servers across 9 categories, every one inheriting the gateway's auth, RBAC, and audit logs. Bring your own and they get the same controls.

Browse the full catalog

04 / Inside the gateway

Distribute AI capabilities from one catalog

The Speakeasy MCP gateway sits between every agent and every system it touches. Your platform team curates the catalog behind it once, access follows your identity provider, employees keep the agents they already use, and adoption shows up as a number.

I don't want anyone using internal UIs or opening SaaS apps. The moment someone shifts a workflow to being AI-native it drops from a two-hour task down to minutes.

Shreyas Kumar

Co-founder, Fermat

Fermat

05 / How it works

From pilot to company-wide in four steps

The order matters: identity first, then the catalog, then the fleet, then the numbers. Each step is a page in the product rather than a project.

  1. 01

    Connect your IdP

    SSO and Directory Sync from Okta, Entra, or any SAML or OIDC provider. Teams and roles arrive already defined.

  2. 02

    Curate the catalog

    Pick servers from the catalog or add your own. Assign each by team and role. Security reviews the plane once, not every server.

  3. 03

    Instrument the fleet

    The device agent ships through your MDM and configures Claude Code, Cursor, Codex, and the rest with the org's MCP servers, Skills, and hooks.

  4. 04

    Measure and expand

    Watch enrollment, sessions, and cost by team. Move the next team onto the catalog with the numbers from the last one.

Results

When AI is governed, adoption grows

Numbers from teams running the control plane in production: full-company rollouts, MCP servers shipped in days, and usage that grows because governance stops being per-project work.

Read the Fermat story

100%

Employee AI rollout

Fermatrolled out governed AI to the whole company in two days.

200%

MCP usage growth

PlanetScalecustomers are now agent-based.

200+

MCP servers rolled out in 30 days

MoonPaywent from proof of concept to company-wide in a month.

Questions

Which AI clients does the gateway support?
Any MCP client. Claude, Claude Code, Cursor, ChatGPT, Codex, Copilot, Gemini, and custom agents connect through one URL with OAuth 2.1, PKCE, and dynamic client registration, so modern clients work without custom integration.
Do teams have to give up the tools they already use?
No. Existing clients connect to the gateway, and the device agent configures them centrally with the organization's MCP servers, Skills, and plugins. The change for an employee is that their agent now reaches the systems it needs, under their own identity.
How does a new MCP server get approved?
Add it from the catalog or register your own, assign it by team and role, and it inherits the gateway's authentication, policy, and audit logging on day one. Servers that show up in agent traffic without being registered are blocked by default and wait in the review queue with the evidence attached, so the platform team approves or denies with the usage in front of them.
What about employees on personal AI accounts?
The device agent keeps the organization's plugins and MCP configuration in place on every laptop, so an agent running on a personal plan still reaches company systems through the gateway, under the employee's identity, with the same policy applied. Coverage stops depending on which licenses the company centrally manages.
How long does a rollout take?
Fermat rolled out centrally managed MCP access to the whole team in two days. MoonPay went from proof of concept to a company-wide rollout of 200+ MCP servers in 30 days. Read the MoonPay story.

Go deeper

ConnectAgent identityDistributeMCP gatewaySecureAgent securityObserveAI usage telemetry

AI everywhere.

Control here.

Talk to usRead the guide