← Back to Discover

Speakeasy

Enterprise layer between AI agents and company systems that connects them through an MCP catalog, enforces access and security policy, and records every tool call for audit and usage reporting.

MCP serversObservability & evals
Visit website ↗

What it does

Speakeasy's AI control plane is a managed or self-hosted layer that sits between AI agents and the systems they reach, such as SaaS tools, internal APIs and databases. The site describes four jobs for it: connect, secure, distribute and observe. Connecting works through a central catalog of MCP servers. Pre-built integrations cover tools such as Salesforce, Slack and HubSpot, and internal APIs can become managed MCP servers from an OpenAPI spec. Access follows the roles and teams in an identity provider such as Okta, Microsoft Entra ID or any SAML or OIDC provider, and permissions can be scoped down to a single tool. On the security side, prompts, responses and agent actions are inspected as they pass through. The site says PII and credentials are blocked or redacted, prompt injection and shadow MCPs are detected, and alerts can go to Slack, PagerDuty or a SIEM. Observability covers usage by team, tool and user, adoption against rollout targets, and an audit trail of tool calls. It is described as working with MCP-compatible clients including Claude, ChatGPT, Cursor, Copilot and Codex, as well as internal agents. The page lists SOC 2 Type II and ISO 27001, and says the data plane can be deployed in your own VPC. Prices are not shown on this page; it directs visitors to talk to the team.

Capabilities

  • Central catalog of MCP servers and tools, provisioned by team and role
  • Pre-built SaaS integrations, plus managed MCP servers generated from an OpenAPI spec
  • Identity-provider integration with Okta, Entra ID, Auth0 and any SAML or OIDC provider
  • Real-time blocking of PII and secrets, with prompt-injection and shadow-MCP detection
  • Usage and adoption reporting by team, tool and user, with a full audit trail
  • Managed cloud or self-hosted deployment

When to use it

  • Rolling out AI agents across a company with one approved set of tools
  • Giving security teams a record of what agents read and change
  • Turning internal APIs into MCP servers without custom integration work for each
  • Measuring which AI tools teams actually adopt