Boomi adds an MCP Registry to its integration and API control plane
- Enterprise Integration
- API Management
- AI
What happened: Boomi's May 2026 platform release introduced an MCP Registry for centralized AI agent tool governance, alongside broader gateway discovery, Kong Konnect discovery, a discovery agent SDK, and deeper Cloud API Management asset visibility.
Why it matters: This is another integration-platform vendor treating MCP tools as governed enterprise assets, not scattered developer endpoints. It maps directly to the capability categories our consultants already use for API discovery, lifecycle, consumer onboarding, analytics, and policy enforcement.
Enterprise adoption impact: Customers with heterogeneous gateway estates will start asking whether one control plane can inventory traditional APIs, MCP servers, agent tools, subscriptions, and ownership metadata. This creates a concrete platform-rationalization conversation.
Watchpoint: Test whether MCP registry entries carry enough operational metadata: owner, environment, authentication model, risk class, allowed consumers, audit trail, deprecation path, and telemetry export.
AWS turns SAP OData services into MCP tools for agents
- AI
- Enterprise Integration
- Cloud Architecture
What happened: AWS described the AWS for SAP MCP Server on Amazon Bedrock AgentCore, exposing SAP business data and processes as MCP tools built on SAP OData APIs, with Bedrock AgentCore Runtime handling hosting, session isolation, connectivity, and authorization.
Why it matters: Agent integration is moving toward packaged bridges into core systems of record. That is attractive, but also risky: SAP actions are business-critical, stateful, and permission-sensitive.
Enterprise adoption impact: SAP-heavy clients may see this as a shortcut from AI experiments to operational workflows. Architecture reviews need to separate read-only assistance, guided workflow proposals, and direct mutations of finance, procurement, logistics, or master-data objects.
Watchpoint: Ask whether every tool call can be tied back to a human or service identity, whether SAP authorization is preserved end to end, and whether compensating controls exist for failed or partially completed agent actions.
Enterprise agent controls are becoming workspace administration work
- AI
- Automation Platforms
- Enterprise IT Architecture
What happened: OpenAI's Enterprise and Edu release notes for May 28 and May 29 added workspace-agent controls such as model and reasoning effort settings, role-based publishing permissions, guided setup, speech output, smarter Slack thread replies, and Codex updates for Windows computer use and GitHub Enterprise app templates.
Why it matters: Agents are becoming managed workspace objects. That shifts governance from isolated prompt engineering to admin policy: who may publish agents, which apps they can access, which model profile they use, and where execution logs are available.
Enterprise adoption impact: Clients will need a catalog of approved agents and connector scopes, much like they already manage enterprise apps and API consumers. Slack-thread behavior also means collaboration tools become part of the agent runtime surface.
Watchpoint: Build a lightweight control checklist for workspace agents: publisher role, data sources, app access, run logs, escalation path, offboarding, and region or platform availability constraints.
MCP security research keeps pointing at identity and admission control gaps
- AI
- Security
- API Management
What happened: Recent MCP research highlights authentication weaknesses in real-world remote MCP servers and proposes stronger admission patterns for tool servers. One measurement study of OAuth-enabled MCP servers found recurring dynamic client registration and delegated authorization flaws.
Why it matters: The market is racing to expose tools to agents faster than security models are maturing. MCP governance cannot stop at cataloging; it must decide which server, tool, method, data scope, and identity context are allowed for each agent.
Enterprise adoption impact: Security teams will expect agent gateways to integrate with identity providers, token policies, supply-chain checks, runtime logging, and approval workflows. This strengthens the case for joint API management, IAM, and security architecture reviews.
Watchpoint: Treat every remote MCP server like a third-party API plus executable workflow surface. Require threat modeling before onboarding high-impact tools.