Same Governance, Different Boundary
Managed and self-hosted options differ by operating boundary, not by governance intent.
Why it mattersTeams can reason about deployment without relearning the control model.
Same Governance Kit
Identity, policy, credentials, sessions, and audit remain the core kit.
Why it mattersThe same primitives should govern agent access wherever the gateway runs.
Managed Runtime, Approved Endpoint
The managed option runs DAC-hosted control plane and runtime paths for approved customer HTTPS endpoints.
Why it mattersTeams can try governed calls quickly while keeping endpoint registration, policy, credentials, and audit explicit.
Customer Endpoint Allowlist
Managed access starts with a customer-owned HTTPS API or HTTPS MCP endpoint that allowlists DAC egress.
Why it mattersThe first managed path is deliberate endpoint access, not broad private-network reachability.
Managed Scope Boundary
Managed does not mean arbitrary managed MCP server hosting or generic proxying.
Why it mattersScope clarity keeps V1 claims credible and helps security teams know exactly what traffic can run.
Self-Hosted Everything Inside
In self-hosted mode, control plane, data plane, registry, policy, and audit all live inside the customer boundary.
Why it mattersSome customers need operational ownership more than managed convenience.
Customer-Owned Substrate
Self-hosted deployment can attach to customer-owned Postgres, cache, IdP, secrets, and SIEM systems.
Why it mattersEnterprise adoption often depends on fitting into existing platform responsibilities.
No Outbound Runtime Dependency
Normal operation can stay local, with telemetry and export under customer control.
Why it mattersTrust-boundary decisions are often about runtime dependency, not only data location.
The Call Still Gets Governed
Calls still authenticate, evaluate policy, broker credentials, route privately, and emit audit evidence.
Why it mattersDeployment option should not weaken the runtime governance path.
API Adapter Stays Bounded
The adapter still exposes selected operations through local checks and upstream status receipts.
Why it mattersAPI-to-MCP conversion remains bounded even when the gateway is deployed differently.
Sessions And Revocation Match
Session ID, affinity, drain, revoke, and audit behavior should match across modes.
Why it mattersOperators need predictable controls when they move from pilot to production.
Choose By Operating Boundary
The deployment decision starts with speed, endpoint exposure, operational ownership, and runtime dependency.
Why it mattersManaged and self-hosted are choices about where work runs and who owns the boundary.
Deployment model
Prove governed MCP inside your trust boundary
We are looking for teams who want to prove governed MCP inside their real trust boundary.
The first pilot should choose the boundary deliberately: managed where DAC-hosted runtime and allowlisted endpoints are acceptable, self-hosted where customer-owned runtime and control plane are required. Then connect one approved endpoint or server, run one policy test, and inspect one audit proof.
The goal is not to debate deployment in abstract. It is to test the boundary with a real workflow.
Discuss your deployment