LoginSign Up
← Back to all sparks
S

Speakeasy

DEVOPS
Velocity10.0

Speakeasy becomes the enterprise control plane for MCP server access and AI tool governance.

mcp-governanceenterprise-access-controlshadow-aiai-securityagent-infrastructure
◆Current state
Speakeasy has pivoted from an SDK/API generation tool into a full governance platform for Model Context Protocol infrastructure. In the past week, the product shipped per-server access control pages, per-tool permission granularity, killswitch management, credential verification for remote sessions, and a breaking refactor to risk policy scoping. The pace of feature delivery across v1.21–v1.25 is high, and the product surface has grown substantially: shadow AI detection, OTLP risk export, catalog management, and editable gateway instructions are all now part of the same control plane.
◆Where it's heading
Every release deepens the enterprise governance story: finer-grained access controls, trusted issuer chains for enterprise IdP integration, and precise billing instrumentation from exact meter readings. The direction is clear—Speakeasy is positioning itself as the security and compliance layer for organizations deploying AI agents at scale. The API-breaking risk policy refactor signals the team is willing to clean house to reach a coherent model rather than accumulate overlapping scope mechanisms.
◆Prediction
The next move is likely deeper audit and compliance tooling—possibly signed event logs, per-user MCP usage reports meeting enterprise security requirements, or wider IdP federation support beyond the current OIDC/RFC 8414 implementation.

◆Recent moves

  1. 1d ago

    The MCP catalog on the MCP page, editable gateway instructions, and a usage explorer backed by exact meter readings

    The MCP catalog is now surfaced directly on the MCP page alongside Servers and Deployments, with command palette integration for discovery. Gateway instructions—the guidance sent to every connected client—are now editable, and the billing explorer reads exact daily meter totals rather than estimates. These UX completions make the platform less opaque to operators managing multiple MCP servers.

  2. 2d ago

    Shadow MCP approvals now bound distribution, trusted issuer links for enterprise authorization, and a fail-open fix in realtime enforcement

    Shadow MCP approvals now scope to the audience they were granted for—approval to distribute to a subset can no longer be leveraged to reach a wider one. Enterprise org admins can link trusted IdP issuers for user session authorization. Both tighten the security perimeter on MCP distribution without adding new surface area.

  3. 4d ago

    Risk policies scope only through detection scopes, and MCP requests no longer trigger toolset indexing

    Risk policies now scope exclusively through detection scopes, dropping three legacy policy-level fields in a breaking API change. MCP request paths no longer trigger toolset indexing—indexing moves to a background sweep, removing a latency spike from the hot path. The breaking change is painful for existing API callers but rationalizes a model that had accumulated multiple overlapping scope mechanisms.

  4. 4d ago

    Know whether a remote session's stored credential still works before anyone dispatches a tool call

    Remote session cards now show credential validity—valid, rejected, or unknown—updated automatically on connect and verifiable on demand via a dry-run dispatch. This closes an operational blind spot where a stored credential could silently fail at tool-call time, surfacing the problem before any user action triggers it.

  5. 6d ago

    Edit an MCP server's access one scope at a time, and manage user session issuers for the whole organization

    MCP server access editing gains per-scope granularity: connect, view, and manage can now be adjusted independently per principal, allowing a role with org-wide server access to be narrowed for a single server without touching the role definition globally. Organization-level management of user session issuers rounds out the access model.

  6. 7d ago

    Point an MCP server at your identity provider's issuer and see who each connection belongs to

    External OAuth backed by IdP discovery rather than manually entered credentials is a requirement for enterprise environments where identity providers rotate configuration. The 'Connected as' display derived from ID token verification closes the gap between 'a connection exists' and 'we know whose it is' — which is the audit evidence compliance teams need when reviewing who accessed MCP tools.