Cloudflare AI Gateway
Choose it when Cloudflare should own the edge layer.
You want managed analytics, caching, rate limits, retries and fallback close to Cloudflare’s network, with MCP access integrated into Cloudflare identity.
← All comparisons
Independent decision brief
DVARA or Cloudflare AI Gateway?
Cloudflare puts model traffic and MCP access on managed edge services. DVARA puts the governance runtime inside your environment. The decision starts with that boundary, not a feature count.
Start with fit
Which operating model matches yours?
A useful comparison should help you eliminate the wrong option quickly.
DVARA
Choose it when the governance boundary must be yours.
You need policy, data handling, spend controls and tamper-evident evidence inside your own environment across model, MCP and A2A traffic.
Four decisions
Compare the boundary, not the checklist
Each row explains the tradeoff and the question to take into a pilot.
Where can the control point run?
This is the clearest architectural difference.
Cloudflare AI Gateway
Cloudflare AI Gateway is a managed Cloudflare service. Requests then reach Workers AI or the third-party model provider selected by the application.
DVARA
DVARA runs in your environment and sends requests only to the model or tool destinations you configure.
Decision: Choose Cloudflare for an edge-managed service. Choose DVARA when the governance runtime and its stored evidence must remain in your infrastructure.
What should happen to model calls?
Cloudflare leads with traffic operations; DVARA leads with governance decisions.
Cloudflare AI Gateway
AI Gateway documents analytics, logging, caching, rate limits, retries, model fallback and optional unified billing.
DVARA
The DVARA LLM Gateway applies workspace policy, PII controls, spend enforcement and audit evidence across supported providers.
Decision: Compare the exact request-path controls you need, including payload retention, policy defaults and which party operates the gateway.
How should MCP access work?
Cloudflare’s MCP portal is not the same surface as AI Gateway.
Cloudflare AI Gateway
MCP server portals centralize servers behind one endpoint and add Cloudflare Access, OAuth, server selection and portal policies.
DVARA
Enterprise deployments add MCP argument policy, PII checks, approval gates and loop detection, with A2A governance in a separate plane.
Decision: Choose based on whether the problem is secure MCP access or runtime governance of tool arguments, approvals and agent behavior.
What evidence do you need?
Edge logs and governance evidence answer different questions.
Cloudflare AI Gateway
Cloudflare provides gateway analytics and logs for managed traffic, subject to the selected service configuration and retention.
DVARA
DVARA records policy decisions and runtime events in a tamper-evident audit trail operated with the rest of the deployment.
Decision: Confirm who stores payloads, how long evidence remains available and whether an auditor can verify why a request was allowed or refused.
Verified details
Facts that change the decision
These details are easy to lose in a broad feature matrix.
- Managed boundary
- Cloudflare AI Gateway runs as a Cloudflare-managed edge service.
- Model operations
- The service documents caching, rate limiting, retries, fallback, analytics and logging.
- Guardrails and DLP
- AI Gateway documents guardrails and data loss prevention on model traffic.
- Provider path
- Requests can reach Workers AI or supported third-party model providers.
- MCP portals
- Cloudflare portals aggregate MCP servers and apply Access, OAuth and portal policies.
- Separate surfaces
- Model traffic and MCP portal governance should not be described as one undifferentiated product.
- DVARA deployment
- DVARA’s governance runtime and evidence store run in the customer environment.
Bottom line
Managed edge or customer-operated governance?
Cloudflare is a strong fit when managed edge routing, caching and Cloudflare-native MCP access are the priority. DVARA is the clearer fit when policy decisions and governance evidence must remain under customer operation across model, tool and agent traffic.