TL;DR: Anthropic’s acquisition of Stainless signals that generating SDKs and MCP servers is becoming routine, but governing agent-to-tool and agent-to-model traffic remains the harder enterprise problem, according to Kong. Runtime policy, telemetry, and auditability now define whether AI connectivity is deployable at scale.
At a glance
What this is: This is Kong’s analysis of Anthropic’s acquisition of Stainless, which it uses to argue that generating SDKs and MCP servers is getting easier while governing AI connectivity in production remains the harder problem.
Why it matters: It matters because IAM, platform, and security teams need runtime controls for who and what can call tools, how those calls are logged, and whether agentic traffic can be audited and costed across models and applications.
Context
The core issue is AI connectivity governance: the controls that decide who or what can call tools, under what conditions, and with what proof for audit and finance. Generating an SDK or MCP server is only the build-time step; it does not govern runtime access, cost, or traceability.
Kong’s article frames Anthropic’s acquisition of Stainless as evidence that connector generation is moving toward commodity status. The enterprise problem is what happens after those connectors exist: policy enforcement, telemetry capture, audit trails, and neutral control across multiple models and agent frameworks.
This is a governance story, not just a tooling story. Once agents can reach more internal systems, the control plane has to keep up with agent-to-tool and agent-to-model traffic, or enterprises end up with reachability faster than accountability.
Key questions
Q: How should teams govern AI agents once connector generation becomes easy?
A: Teams should govern AI agents at runtime, not at generation time. The real control surface is policy enforcement, telemetry, and auditability around every tool call, model call, and event-triggered action. If governance starts only after a connector is generated, the enterprise has already ceded control of access, cost, and traceability.
Q: Why does multi-model AI make connectivity governance harder?
A: Because policy has to be consistent across models that serve different business needs, trust levels, and hosting patterns. If governance is bound to one vendor’s runtime, the enterprise cannot prove that the same access rules, logs, and cost controls apply everywhere. Neutrality becomes a control requirement, not a preference.
Q: What breaks when MCP servers are generated without runtime controls?
A: Call volume grows faster than oversight. Agents can reach more tools, but the organisation loses the ability to explain which model invoked which service, whether the call was authorised, and how correlated failures or spend spikes should be attributed. The result is connector sprawl without operational accountability.
Q: Should enterprises separate connector generation from governance layers?
A: Yes. Connector generation should accelerate developer productivity, while governance should remain a distinct control plane that applies policy, records activity, and handles audit and cost attribution. Collapsing those layers creates an environment where reachability improves faster than control.
Technical breakdown
Generation-to-governance gap in AI connectivity
Spec-to-SDK and spec-to-MCP generation shortens the path from an API description to a callable interface for agents. That helps developers, but it only covers the build-time layer. The governance problem begins when those generated interfaces are live, because policy has to follow every call, every user context, and every downstream dependency. Runtime control is therefore a separate architecture layer, not an extension of code generation.
Practical implication: Treat generated connectors as an input to governance, not as the governance layer itself.
Why agent-to-model traffic needs neutral policy enforcement
Agentic systems do not just call tools. They also depend on model access, event triggers, and shared memory, which means the runtime has to govern multiple communication patterns at once. If the control plane is tied too tightly to one model vendor, it cannot provide consistent policy, telemetry, or auditability across a multi-model estate. Neutrality is not a preference here; it is an architectural requirement for enterprise scale.
Practical implication: Design policy, logging, and metering so they work across Anthropic, OpenAI, Google, AWS, and self-hosted models alike.
Why MCP server generation is not the same as operational control
MCP server scaffolding lowers the friction of exposing tools to agents, but it does not answer authorization, rate limiting, or correlated failure handling. When many agents call many generated interfaces, the enterprise also needs observability into which model invoked which tool on whose behalf and at what cost. Without that layer, connector sprawl becomes a governance blind spot rather than a productivity gain.
Practical implication: Place enforcement, observability, and cost attribution in the path of every agent-to-tool call before broad rollout.
NHI Mgmt Group analysis
Connector generation is becoming routine, but connector governance is the enterprise gap. Turning an OpenAPI spec into an SDK or MCP server is rapidly becoming commoditised. The harder question is not whether an agent can reach a tool, but whether the organisation can prove who authorised that reach, under what policy, and with what audit trail. The generation-to-governance gap is now a category-level problem, not an implementation detail.
Runtime AI connectivity is a separate control plane from model selection. Enterprises will continue to run multiple models for different workloads, data sensitivities, and business units. That means the governance layer must sit above model choice and treat each endpoint as a peer subject to the same policy, telemetry, and accounting rules. A model-vendor-native control plane cannot fully satisfy that requirement in a multi-model estate.
AI connectivity is the right named concept for this architectural split. Build-time connector creation and runtime control are not substitutes, and conflating them will produce overexposed agents with under-governed access paths. The practical consequence is that security, platform, and finance teams need one shared view of call paths, spend, and policy violations. That is where enterprise-ready agentic systems are won or lost.
The neutrality requirement will shape the market more than connector syntax. As model vendors pull more connectivity features into their own stacks, enterprises will still need a control layer that works across vendors and frameworks. That pushes the market toward neutral enforcement, audit, and metering rather than one-off developer convenience. The practitioner takeaway is to evaluate every agent runtime on cross-model governance, not just on integration speed.
What this signals
AI connectivity will become a governance category of its own. Enterprises are already moving past proof-of-concept agents and into environments where tool reach, model access, and memory access all need the same policy model. The organisations that separate generation from runtime governance will scale faster because they can see, explain, and constrain agent behaviour in production.
Runtime neutrality is now a design requirement. The more enterprises mix Anthropic, OpenAI, Google, AWS, and self-hosted models, the less viable it becomes to anchor control in any one vendor stack. Security and platform teams should assume that the control plane must outlive individual model choices and still preserve the same audit and cost signals across them.
For practitioners
- Map the runtime control plane Separate build-time connector generation from the runtime layer that enforces policy, logs calls, and attributes cost across agents and models.
- Enforce neutral access policy Apply the same authorisation, rate limiting, and audit rules to every model endpoint so no vendor-specific stack becomes the only governance boundary.
- Instrument tool-call telemetry Capture which model called which tool, on behalf of which user or workflow, and preserve that record long enough for audit and finance review.
- Govern event and memory paths together Extend policy to agent triggers, event brokers, and shared memory so tool access is not the only governed communication channel.
Key takeaways
- AI connector generation is maturing quickly, but production risk shifts to the runtime layer where access, policy, and auditability are enforced.
- The central governance problem is not whether agents can reach APIs, but whether enterprises can explain and control every tool and model call.
- Practitioners should separate build-time acceleration from runtime governance so agentic systems remain observable, accountable, and vendor-neutral.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Agentic AI Top 10, MITRE ATT&CK and OWASP API Security Top 10 address the attack and risk surface, while NIST AI RMF and NIST CSF 2.0 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| OWASP Agentic AI Top 10 | ASI03 — Identity & Privilege Abuse | Agent-to-tool governance and auditability sit directly in the agent privilege layer. |
| Recommendation — Apply ASI03 to govern agent authorisation, traceability, and privilege boundaries at runtime. | ||
| NIST AI RMF | GOVERN — AI Governance and Accountability | The article is fundamentally about who governs AI connectivity across models and tools. |
| Recommendation — Establish governance and accountability for AI connectivity before scaling agent deployments. | ||
| NIST CSF 2.0 | PR.AA-05 — Access Permissions, Entitlements and Authorizations | The runtime control plane must manage who can call tools and models under policy. |
| Recommendation — Enforce entitlement checks at runtime for every agent, tool, and model interaction. | ||
| MITRE ATT&CK | TA0006; TA0008 — Credential Access; Lateral Movement | Uncontrolled agent connectivity broadens the paths adversaries could use for access and movement. |
| Recommendation — Map agent tool exposure to TA0006 and TA0008 to prioritise control points that limit reach. | ||
| OWASP API Security Top 10 | API5 — Broken Function Level Authorization | Agent-driven API use still depends on function-level authorisation staying intact. |
| Recommendation — Use API5 controls to prevent agents from invoking functions they are not authorised to use. | ||
Key terms
- AI Connectivity: AI connectivity is the enterprise control layer for governing how applications, users, and agents reach AI services. It combines routing, security, observability, and policy enforcement so AI interactions can be managed consistently across providers and use cases.
- Governance Gap: A governance gap is the distance between knowing an asset exists and being able to enforce policy on it. In identity programmes, it appears when discovery, review, and enforcement are split across different tools or teams, leaving access partially visible but not truly controlled.
- Runtime Governance: Runtime governance is the set of controls that verify what a system or agent is actually doing after deployment. It combines monitoring, authorization checks, and access validation so teams can detect drift, misuse, or excessive privilege in motion rather than assuming build-time policy still holds.
- Model-Neutral Control Plane: A governance layer that applies the same policy, logging, and cost rules across multiple AI models and hosting patterns. For agentic systems, neutrality is required when no single vendor should become the sole trust boundary for enterprise traffic.
Deepen your knowledge
NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an IAM programme, it is worth exploring.
Published by the NHIMG editorial team on June 23, 2026.
Updated on October 11, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org