Speakeasy is one of a small number of companies building infrastructure specifically for the agent era rather than retrofitting existing security tools. Its control plane governs MCP servers, agent identities, and tool calls, which puts it directly in the path of how enterprises deploy agents at scale. The company's open-source work is substantial: its GitHub organization has 177 repositories, including the Gram stack for connecting and securing agents, an OpenAPI toolkit in Go, and a coding agent runtime called Kit. It is a member of the Agentic AI Foundation and has partnered with Google on open-sourcing developer and agent experience tooling.
For anyone building agents that need to run inside a regulated enterprise, Speakeasy is worth understanding. The company is betting that identity, policy enforcement, and audit trails become table stakes for agent deployment, and that the layer belongs on the endpoint rather than in the model or the API gateway. Whether that bet pays off depends on how enterprises actually architect their agent stacks, but the customer numbers from MoonPay, Fermat, and PlanetScale suggest the approach is finding buyers.
Speakeasy started in 2021 as a developer tooling company. Its first product generated type-safe, idiomatic SDKs from OpenAPI specs, handling OAuth 2.0, retries, pagination, and webhook verification so API teams did not have to hand-write client libraries in ten languages. That business still runs, and its customers include Vercel, Cloudflare, Mistral, Honeycomb, and PlanetScale. The company raised an $11M pre-seed and seed in 2023, led by GV and Quiet Capital, then a $15M Series A in October 2024 led by FPV Ventures.
After MCP launched, Speakeasy took the same infrastructure it had built for API developer experience and pointed it at a different problem: enterprises rolling out AI agents with no way to see what those agents were doing or what they could reach.
The product catalogs every AI agent, MCP server, and Skill in use across a company. Each agent gets its own identity from the customer's existing identity provider, whether that is Okta, Microsoft Entra ID, Auth0, or any SAML/OIDC provider, so access follows the teams and roles already defined there. Speakeasy deploys through the MDM a company already runs, Jamf, Intune, or Iru, which means coverage is fleet-wide from day one rather than per-team setup.
From there, every prompt, response, and tool call is checked against policy in real time. Allowed actions pass through to internal APIs, SaaS tools, and MCP servers. Violations stop at the control plane. Every allow-or-deny decision lands in an audit trail. The company says it blocks prompt injection, PII exposure, and leaked credentials before they leave the perimeter.
MoonPay governs 60,000 agent sessions per week through the platform. Fermat rolled out governed AI to its entire company. PlanetScale reports 200% growth in MCP usage after adopting it. The company publishes a dashboard view showing total spend, agent sessions, tool calls, and risk findings, which is the kind of telemetry security teams need to justify the spend.
Deployment is fast by enterprise standards. Speakeasy says most teams are running within a day because access inherits from SSO groups. Self-hosted and on-premises options exist for teams that cannot send data to third-party infrastructure.
The control plane is built to produce evidence for EU AI Act logging and human oversight requirements, ISO 42001 AI management system controls, and NIST AI RMF govern-map-measure-manage functions. Speakeasy holds SOC 2 Type II, ISO 27001, and CSA STAR AI certifications, and is GDPR compliant and HIPAA ready. For enterprises where AI adoption is blocked by legal review rather than engineering, that framing matters more than the technical features.
Speakeasy is competing in a category that barely existed two years ago. The alternatives are MCP gateways, AI security vendors, and the status quo of wiki-page policies that nobody enforces. Its bet is that the control plane belongs at the endpoint, deployed through MDM and wired into the identity provider, rather than at the API gateway or in a separate security tool. The company is based in San Francisco and backed by GV, FPV Ventures, and Quiet Capital, with angels including Datadog president Amit Agarwal and Prisma CEO Søren Bramer Schmidt.
A control plane that catalogs, governs, and enforces policy on every AI agent, MCP server, and Skill in an enterprise.
Generates type-safe, idiomatic SDKs in ten-plus languages from an OpenAPI spec, wired into CI/CD.
Speakeasy is hiring.