Docker open-sources an MCP Gateway
Docker open-sourced an MCP Gateway between AI agents and MCP servers with signature checks, call logs, and secret blocking options.
By Hypler editorial team · Source: Docker.
Original source date: .
Hypler publication: . Hypler updated: .
Estimated reading time: 4 minutes.
Topics: Docker, MCP, security, gateway

The announcement proposed a central boundary
Docker announced its open-source MCP Gateway on July 9, 2025 as a process between AI agents and multiple MCP servers. The launch article described Docker Compose compatibility and configuration for selected servers and tools. It listed initial interceptors for image-signature verification, inbound and outbound secret blocking, and call logging, and said the gateway was available through Docker Desktop and as an open-source project. The source presents an architecture for discovery, configuration, execution, and security controls around MCP workloads. It does not establish that a particular server is trustworthy or a particular call appropriate.
Policy must be reviewed configuration
Define allowed clients, servers, transports, tool names, argument constraints, rate limits, authentication requirements, log destinations, and failure handling in versioned configuration. Separate tenant, environment, and sensitivity policies so a developer gateway cannot inherit broader credentials or a larger server list. Expose only the server and tool subset a workflow requires. A policy diff should make additions visible: a new server, widened argument, new outbound host, or modified interceptor. Those details are the operational meaning of centralized enforcement, and they give reviewers something more durable than an assertion that traffic passes through a gateway.
Interceptors supplement the controls at either end
Signature verification, secret scanning, and logging address important supply-chain and transport conditions. They work alongside server-side authorization, schema validation, target-object checks, and application approval rules. Capture an interceptor decision with reason code, configuration version, and request identifier. Keep the server result separate from the gateway result, allowing operators to distinguish a blocked call, rejected request, unavailable server, and executed request with uncertain outcome. That distinction matters during incident review; a single log line should not be expected to explain every authority and execution decision along the path.
A two-server gateway illustrates constrained use
Imagine a hypothetical support gateway for one agent client. It exposes a public-status lookup server and a read-only knowledge-base server on a named transport. Policy pins the two server images by digest, allows only status and search tools, blocks payloads matching configured secret patterns, and writes structured call records to retained storage. The client gathers cited status and knowledge excerpts for a draft response. Customer updates, ticket changes, and message sending stay in separate workflows with their own approval paths. The gateway makes one narrow tool surface visible and testable instead of confusing access to information with authority to act.
Logs need context without becoming a data sink
Record correlation identifier, authenticated client, policy revision, server digest, selected tool, argument classification, interceptor decisions, timestamps, response category, and evidence reference. Redact secrets and sensitive data under a defined policy rather than relying on improvised string filters. Link retries to the originating request so duplicates and ambiguous network failures can be reconciled. Monitor denied calls, secret blocks, unknown tool requests, policy mismatches, transport errors, and long-running calls. Those signals reveal drift while the relevant configuration is still inspectable and keep observability from becoming unbounded capture of task content.
Failure behavior and capacity are control behavior
Start with a fixed client and small server set. Test allowed requests, blocked tools, invalid arguments, expired credentials, untrusted output, signature failures, secret-pattern matches, restarts, and partial responses. Verify that intended topology does not permit a direct bypass. Set timeouts, concurrency limits, connection pools, request-size limits, and backpressure, then exercise them under load. A saturated gateway must return a visible classified failure, not silently drop audit events or forward partially validated work. Keep rollback configuration for a changed policy or server. Re-run the suite after gateway, server image, interceptor, or transport changes.
Hypler analysis
Hypler’s view
Centralized enforcement is useful defense in depth, not a verdict that an action should occur. A gateway becomes a real control plane only when its allowlists, policy revisions, records, capacity, and high-impact approvals are explicit.
Hypler technical analysis
Engineering notes
A gateway can centralize the boundary between agents and multiple MCP servers, allowing policy and evidence to be applied where requests fan out. Signature verification, logging, and secret blocking are controls with different failure modes; they need explicit allowlists, decision records, and review of what remains permitted after the gateway passes a call.
What to inspect
- Tool policy
- Inspect allowlists by client, server, operation, argument class, and environment; test default denial and a request that reaches an unauthorized server.
- Secret boundary
- Test credentials in tool arguments, returned data, logs, and error strings; inspect blocking behavior plus the secure credential reference used by an allowed call.
- Audit chain
- Trace a high-impact request through gateway decision, upstream tool result, retries, and final outcome; verify records remain correlated and retained as required.
Scope and limits. A gateway is a defense layer, not a proof that a server is trustworthy or an individual call is appropriate. Human accountability and scope review remain separate requirements.
Terms and definitions
- MCP Gateway
- A service boundary that routes MCP traffic and can apply discovery, authentication, authorization, and policy controls.
- Least Privilege
- A security principle that grants only the minimum access needed for a specific task.
- Supply-chain Security
- Controls that reduce risk from compromised or untrusted software, dependencies, build systems, models, and hardware inputs.
- Execution Evidence
- Records that show whether an approved action was attempted, what inputs and targets it used, and what result occurred.
- Docker adds agent-focused building blocks to Compose
Docker announced agent-focused building blocks for Docker Compose, describing a compose.yaml-based way to define models, agents, and MCP-compatible tools together.
- Docker announces an MCP Catalog and Toolkit
Docker announced the Docker MCP Catalog and MCP Toolkit for discovering, launching, and managing Model Context Protocol servers.
- OpenAI's containment report separates agent permissions from network isolation
OpenAI reports stronger containment for frontier-model research. Hypler examines network isolation, stop-system evidence and controls smaller teams can inspect.

Original source
Historical source briefing with Hypler engineering analysis. Source details reflect the original publication date; implementation examples are hypothetical. Technical analysis expanded October 24, 2026; this does not change the original source event or establish a new source verification. Editorial prose revised October 26, 2026; original event, publication and source-check dates are retained.