# The only enterprise-grade. MCP Gateway. Build a governed golden path between your agents and your data. Deploy an MCP gateway that provisions access from your IdP, normalizes authentication, and generates a full audit trail of every tool call. [Deploy for free](https://ai.speakeasy.com/sign-up) ## The closed loop MCP management solution Speakeasy is the only MCP gateway embedded in the agent loop. The platform captures every server agents connect to, approved or not. Admins promote the ones worth keeping into the official registry, roles from your IdP decide who gets them. ### 1. Discover your shadow MCP Employees across every function connect MCP servers to their AI clients long before anyone asks for approval. The gateway surfaces every server in use, who is using it, and which tools it exposes, so the inventory stays current without a survey. - Shadow MCP surfaced automatically - Usage by employee, client, and tool - Unsanctioned servers blocked by default ### 2. Promote to your registry Admins stop being the approval queue. Promote a discovered server into the official registry in one step, filtered to the tools you approve, with auth, policy, and logging attached on day one. - One-step promote to the registry - Tool-level allowlist per server - Bring your own servers, same path ### 3. Distribute to your users Roles from your IdP decide who gets which servers. Assign a server to a group in Okta or Entra and it appears in every member's Claude, Cursor, and ChatGPT. Backend agents reach the same servers through one endpoint with identities of their own. - Provisioned by role from your IdP - Native connectors for employees, one endpoint for backend agents - Per-team registries, new joiners inherit access ### 4. Govern your usage Every tool call, from any client, passes one control point. Sessions sign in through your IdP with OAuth 2.1, RBAC scopes access down to the tool, prompt injection and PII are caught in flight, and every call lands in one audit log streamed to Datadog, Splunk, or the SIEM you already run. - OAuth 2.1 with CIMD and EMA - RBAC down to the individual tool - One audit log, exported to your SIEM or observability stack ## Connect any MCP server Add GitHub, Stripe, Linear, Notion, and any other MCP. Every MCP gets IdP sign-in, tool-level RBAC, and audit logs on every call. Build your own servers, get the same controls. - [GitHub](/product/mcp-gateway/catalog/github) - [Notion](/product/mcp-gateway/catalog/notion) - [Linear](/product/mcp-gateway/catalog/linear) - [Stripe](/product/mcp-gateway/catalog/stripe) - [PostHog](/product/mcp-gateway/catalog/posthog) - [Supabase](/product/mcp-gateway/catalog/supabase) - [Figma](/product/mcp-gateway/catalog/figma) - [HubSpot](/product/mcp-gateway/catalog/hubspot) - [Context7](/product/mcp-gateway/catalog/context7) - [Zapier](/product/mcp-gateway/catalog/zapier) - [dbt Cloud](/product/mcp-gateway/catalog/dbt-cloud) - [PagerDuty](/product/mcp-gateway/catalog/pagerduty) ## Your identity provider. Your observability. Your network. ### Provisioned by your identity provider Your IdP is the source of truth for permissions, not only identities. Assign a server to an Okta or Entra group and every member gets it in their client. Move someone out of the group and access ends with the next sync. - App assignments serve as a default for who gets which servers - Joiners, movers, and leavers handled by the sync - OAuth 2.1 with CIMD and EMA, even for servers without native-support Works with: Okta, Microsoft Entra ID, Auth0, WorkOS, Google Workspace, Ping Identity, Any SAML / OIDC ### Exported to your observability stack Every tool call is written to one audit log with the employee, client, server, tool, arguments, and result, then streamed to Datadog, Splunk, or any OpenTelemetry collector. - One record per tool call, tied to a named employee - Streams as OpenTelemetry to any collector - Alerts land in the incident workflow you already run Works with: Datadog, Splunk, Grafana, OpenTelemetry, Any OTLP collector ### Served on your private network Run the gateway on your Tailscale tailnet at a stable private hostname and close the public path entirely. Every request carries the Tailscale user, device, and tags it came from, your tailnet ACLs decide who can reach it, and MCP tunnels bring servers inside your network onto the same governed path. - Runs on your tailnet; public hostnames return 403 - Tailscale user, device, and tags on every call - Tunnels for servers that never touch the public internet Works with: Tailscale, Cloudflare, WireGuard, OpenVPN, IP allowlist, MCP tunnels ## Security enforced on every tool call. Once the gateway is the only path between agents and servers, these controls run on every call, for every client, without per-server setup. ### Session quarantine A session that trips a policy is quarantined. Further tool calls are held, security is alerted, and the transcript is one click from the alert. ### Data loss prevention Secrets, personal information, and regulated records are detected in tool arguments and results and redacted at the gateway, before they reach the model and before they leave your systems. ### Prompt injection detection Every request and response is inspected in flight. Instructions smuggled into tool output are flagged before the agent acts on them. [Explore Agent Security](/product/agent-security) ## Support that speeds up development Rollouts move as fast as the answers you get. Support is measured against SLAs, not best effort. These are the current numbers, as of October 2026. - **100% SLA compliance**: Every response-time commitment met, across every support tier. - **23.9m p90 first response**: 90% of support requests answered in under 24 minutes. - **98% Satisfaction rate**: Measured across every resolved support conversation. ## Frequently asked questions ### What is an MCP gateway? An MCP gateway sits in front of every MCP server and turns many ad-hoc tool connections into one governed entry point, where authentication, authorization, and inspection happen consistently. For the full architecture, read what is an MCP gateway? ### How is the MCP Gateway different from running individual MCP servers? Individual MCP servers solve point problems but leave you with N auth surfaces, N audit logs, and N places to enforce policy. The MCP Gateway gives you one URL for every agent, one identity surface, one policy layer, and one audit log. New servers join the catalog and inherit the MCP Gateway's controls instead of being rolled out one by one. ### Which AI agents connect through the MCP Gateway? Claude, Claude Code, ChatGPT, Cursor, Copilot, Codex, and the backend agents you run. In an AI client, the employee signs in through your IdP with OAuth 2.1, even when the upstream MCP server does not support it. Backend agents connect under a workload identity of their own. Each agent is granted a static credential that it uses to request short-lived access tokens. ### What identity providers does the MCP Gateway integrate with? Okta, Microsoft Entra ID, Auth0, WorkOS, Google Workspace, Ping Identity, and any SAML or OIDC provider. Plug your IdP in once at the MCP Gateway and every MCP server behind it inherits your auth, with SCIM and directory sync available where supported. ### How does runtime guardrailing work? Every prompt, response, and tool call is inspected in real time. PII and data exfiltration patterns are actively blocked. Prompt injection and shadow MCPs are passively detected. Alerts integrate with your existing SIEM, observability stack, and incident response workflows so the security team works in the tooling they already use. ### What is a gateway endpoint? A gateway endpoint is one MCP URL that sits in front of a set of MCP servers. Instead of every server's full tool catalog, the agent gets four tools: list what is reachable, describe one server, fetch schemas for the tools it needs, and execute. Tool lists stay small, agents discover tools as they go, and each member server keeps its own auth and access rules. ### Can I bring my own MCP servers? Yes. The MCP Gateway works with any MCP server: pre-built integrations from the Speakeasy catalog, MCP servers you build in-house, and third-party servers you adopt. Every server inherits the MCP Gateway's auth, RBAC, and audit logs, regardless of where it came from. ### What do the audit logs capture? Who called which tool, on which server, with what arguments, against which data, with what result, and when. Audit logs are exportable, queryable, and stream to the SIEM or observability stack your security team already uses, including Datadog and Splunk. ### How do teams roll out the MCP Gateway across an organization? Most teams start by routing one client (often Claude or Cursor) at one MCP server through the MCP Gateway, then expand. Per-team registries let you scope what each team sees, so engineering, sales, and security each get the right catalog without one team blocking another's adoption. [Deploy for free](https://ai.speakeasy.com/sign-up) [Talk to us](/talk-to-us) --- Index of every Speakeasy page for agents: [llms.txt](https://www.speakeasy.com/llms.txt). AI control plane and MCP gateway content in one file: [llms-full.txt](https://www.speakeasy.com/llms-full.txt). Any page is available as markdown by appending `.md` to its URL or by requesting it with `Accept: text/markdown`.