Roadmap

What shipsnext.

The work in progress across the Speakeasy AI control plane, mirrored from our public roadmap. Quarters are what we are planning toward, not commitments. Vote on anything here and we read it.

Request a feature What already shipped

In progress

Being built now, across every pillar.

Planned

Scoped and scheduled for a quarter.

Shipped

Live in the product, with release notes.

01 / Connect

Every agent, every system, one plane

MCP servers, SaaS connectors, and internal APIs provisioned through one catalog and reached through one gateway URL, with the clients your teams already run.

9 items

In progress

Claude in Slack

Connect

Claude Tag support

Govern Claude in Slack like every other client. The same identity, policy, and audit trail apply to every tool call it makes on a user's behalf.

TimelineQ3 2026

StatusIn progress

Vote on the portal

Role plugins · day one

Connect

Automatic MCP distribution

Servers reach the right teams on day one. Role-based plugins land in the default project at onboarding, and anything outside a role's scope goes through an access request instead of a ticket.

TimelineQ4 2026

StatusIn progress

Vote on the portal

ai.speakeasy.com

Connect

Control plane on custom domains

Run the control plane on ai.speakeasy.com and on your own domain. The app becomes host-agnostic and app.getgram.ai is retired.

TimelineQ4 2026

StatusIn progress

Vote on the portal

Stateless · 2026-07-28

Connect

MCP 2026-07-28 spec support

Stateless MCP advertised on every gateway endpoint, with legacy clients and proxy isolation preserved.

TimelineQ4 2026

StatusIn progress

Vote on the portal

Verified publishers

Connect

MCP catalog

A Speakeasy-maintained catalog of MCP servers with verified publishers, tool manifests, and one-click distribution through the gateway. Replaces the third-party registry the catalog pulls from today.

TimelineQ4 2026

StatusIn progress

Vote on the portal Read more

Shipped

GitHub Copilot

Connect

GitHub Copilot support

Copilot runs on the control plane with the same identity, policy, and audit trail as Claude, Cursor, and Codex.

TimelineQ3 2026

StatusShipped

Discuss on the portal Release notes

First-party server · skills

Connect

Platform MCP

A first-party MCP server plus official skills for running the control plane from your agent: register, distribute, and diagnose MCP servers.

TimelineQ3 2026

StatusShipped

Discuss on the portal Release notes

Tailnet ingress

Connect

Private MCP gateway on Tailscale

The gateway served on your tailnet, with the public path closed if you choose. Every request carries Tailscale user, device, and tag identity.

TimelineQ3 2026

StatusShipped

Discuss on the portal Release notes

Outbound only

Connect

MCP tunnels

Connect private, on-prem MCP servers over an outbound-only tunnel, with no inbound ports, governed like every other server.

TimelineQ2 2026

StatusShipped

Discuss on the portal Release notes

02 / Control

Identity and access for agents

Who can use what, under which conditions. Scoped access by team, role, and tool, with the identity provider you already run as the source of truth.

7 items

In progress

XAA · cross app access

Control

Okta Cross App Access

Trust Speakeasy from your Okta tenant and assign applications to it using the permission structure Okta defines for XAA, so the control plane fits the admin model you already run.

TimelineQ3 2026

StatusIn progress

Vote on the portal

SSO · directory sync · app import

Control

Okta-aware onboarding

Connect Okta, turn on SSO and directory sync, and import the applications your organization already assigns, in one guided flow.

TimelineQ3 2026

StatusIn progress

Vote on the portal

Okta · identity chaining

Control

Enterprise managed authorization for outbound MCP

Authorize downstream MCP connections through Okta identity chaining. A user signs in once and every connected service inherits that identity, with no per-service OAuth prompt.

TimelineQ4 2026

StatusIn progress

Vote on the portal

RFC 7523 · SEP-1933

Control

Workload identity federation for MCP servers

Accept platform-issued JWTs from Kubernetes, GitHub Actions, SPIFFE, GCP, and AWS as an RFC 7523 grant on the token endpoint, so workloads reach MCP servers with no Speakeasy-issued secret.

TimelineQ4 2026

StatusIn progress

Vote on the portal

Planned

Revocable · one endpoint

Control

Static credentials for headless agents

A revocable static credential for agents that cannot complete an interactive OAuth flow, scoped to one gateway endpoint.

TimelineQ4 2026

StatusPlanned

Vote on the portal

Shipped

One person · every client

Control

MCP kill switch

Cut one person's MCP clients off from your servers without deprovisioning them.

TimelineQ3 2026

StatusShipped

Discuss on the portal Release notes

CIMD · admission control

Control

Verified MCP clients with CIMD

URL-based client identity and admission control for MCP OAuth. Only clients you have verified get a token.

TimelineQ3 2026

StatusShipped

Discuss on the portal Release notes

03 / Secure

Inspection on every call

Prompts, responses, and tool calls inspected in flight. Policies that block, redact, or ask for review, with every verdict recorded.

6 items

In progress

Inference hooks

Secure

Anthropic Inference Hooks

Speakeasy as the AI security server for Claude Enterprise. Inline verdicts on Claude chat, Claude Code, and Claude Design before the model sees the prompt, reusing the risk policies and classifiers you already configure.

TimelineQ4 2026

StatusIn progress

Vote on the portal Read more

Planned

In flight · by policy

Secure

Tool call redaction

Redact sensitive fields from tool requests and responses in flight, by policy, so the original value never leaves the perimeter.

TimelineQ4 2026

StatusPlanned

Vote on the portal

Warn · continue

Secure

Warn policies for Inference Hooks

Warn-and-continue verdicts on the Inference Hooks path, alongside the block verdicts it supports today.

TimelineQ4 2026

StatusPlanned

Vote on the portal

Human review

Secure

Assisted false-positive review

Flag likely false positives in risk findings for human review, and feed the decisions back into detection.

TimelineNot scheduled

StatusPlanned

Vote on the portal

Shipped

Enforce at the proxy

Secure

LiteLLM integration

Enforce policy where your model traffic already flows, with LiteLLM as the enforcement point.

TimelineQ3 2026

StatusShipped

Discuss on the portal Release notes

Evidence · scoped grant

Secure

MCP approval workflow

A shadow MCP block opens a review with assembled evidence. Approve or deny with a scoped grant and a recorded rationale.

TimelineQ3 2026

StatusShipped

Discuss on the portal Release notes

04 / Observe

The record of what AI actually did

Sessions, tokens, tool calls, and risk findings by team, client, and user, exported to the stack your security and finance teams already run.

8 items

In progress

Query · sessions · metrics

Observe

Analytics API

A declarative query endpoint over sessions, tool calls, and every emitted metric, for the dashboards and reports you build yourself.

TimelineQ3 2026

StatusIn progress

Vote on the portal

Client detection

Observe

Shadow AI monitoring

Extend shadow MCP detection to AI client use on managed devices: which clients run, where, and against which systems, with device agent coverage as the baseline.

TimelineQ3 2026

StatusIn progress

Vote on the portal Read more

Planned

Applied policy · evidence

Observe

Device attestation

Devices report the policy state actually applied, so coverage moves from a heartbeat to verified enforcement, with evidence you can hand to Drata or Vanta.

TimelineQ4 2026

StatusPlanned

Vote on the portal