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.
By Hypler editorial team · Source: Docker.
Original source date: .
Hypler publication: . Hypler updated: .
Estimated reading time: 4 minutes.
Topics: Docker, MCP, tool discovery, containers

The launch described discovery and local controls
Docker announced the Docker MCP Catalog and MCP Toolkit on April 22, 2025. The launch article described a Docker Hub catalog with publisher verification, versioned releases, curated collections, and pull-based distribution for MCP tools. It presented the toolkit as a local way to launch servers, manage credentials and OAuth, expose enabled tools through a gateway, and use a Docker MCP command-line interface. Docker also described memory, network, and disk isolation. These are product and roadmap statements from the launch, not proof that a selected server is appropriate for a particular client or data set.
A catalog entry begins an adoption decision
For every adopted server, retain publisher identity, image reference and digest, declared tools, schemas, required secrets, OAuth scopes, network destinations, persistent volumes, update owner, and retirement plan. Pin the image digest used for development and tests; review a new digest before it reaches another environment. That record connects a catalog choice to operational ownership. A listing and a publisher badge can be useful provenance signals, while digest pinning identifies the exact bytes under review. Neither signal grants a model permission to call the server or establishes the safety of its runtime configuration.
The launch configuration is the actual artifact
Container packaging makes the image, dependencies, and execution settings inspectable together. Review image digest, entrypoint, environment variables, mounts, user identity, resource limits, network policy, and exposed transport as one configuration. Keep fixtures separate from customer and production data. Require a deliberate approval for any added secret, writable mount, outbound destination, or state-changing tool. A useful review asks what the server can receive, where it can send information, and which client may invoke it. That is more informative than treating all catalog tools as interchangeable developer conveniences.
A narrow tool set makes the boundary testable
Consider a hypothetical local development environment that exposes a documentation-search server and a read-only issue-query server to one coding client. The manifest pins both images by digest, grants each only the credential it needs, and permits network access solely to the documented API host. Its gateway enables a named tool list and records invocation with client identity and server version. The client can receive search results and issue metadata; issue creation, source writes, and deployment operations are absent. Catalog discovery has become a reproducible surface whose purpose, effects, and limits can be checked.
Credentials need an operating lifecycle
Keep credentials outside images and source control, inject them only into the server that needs them, and make scope visible in the adoption record. OAuth grants should identify principal, scopes, expiry, revocation path, and the server instance receiving the token. Rotate through a documented process that verifies the replacement and removes the prior value. Diagnostic output must not reveal secret values, authorization headers, or full unredacted arguments. Credential inventory, rotation evidence, and access-denial tests turn a local convenience feature into a practice an owner can actually defend after an incident.
Enablement and removal both require proof
Test approved and denied requests, malformed input, unavailable dependencies, expired credentials, network-policy denial, repeated calls, untrusted text, oversized output, and ambiguous failures. Confirm image, manifest, and logs match the reviewed configuration, retaining results with the digest and catalog revision. Subscribe the owner to upstream releases, advisories, deprecations, and schema changes. Test removal too: disable a server in fixtures and verify clients receive a classified unavailable result, dependent work stops cleanly, credentials are no longer injected, and audit history remains searchable. Hypler's analysis favors a small, evidenced catalog over broad, unowned access. That choice reduces surprise during both operation and retirement.
Hypler analysis
Hypler’s view
A catalog improves discovery while expanding the tool boundary. Publisher verification and versioning help establish provenance, but a team still has to decide whether a server's data access, secrets, network routes, and effects are acceptable.
Hypler technical analysis
Engineering notes
A catalog and local toolkit make MCP servers easier to discover and run, which increases the number of potential tool, credential, container, and network relationships. Distribution provenance can narrow investigation, but authorization requires examining the exact server version, client grant, configuration, and effectful operations in use.
What to inspect
- Artifact provenance
- Inspect publisher identity, selected server version, image or package reference, update policy, and the evidence used to approve that exact artifact.
- Runtime configuration
- Inspect environment-variable sources, mounted volumes, credential references, container isolation, reachable networks, and startup arguments for the local server.
- Client grant
- Record compatible client identity, allowed tools, OAuth or credential scope, permitted data domains, and audit events for failed and successful calls.
Scope and limits. Publisher verification, versions, and curated collections are provenance signals. They do not certify a server's purpose, secrets handling, network behavior, or suitability for a specific workflow.
Terms and definitions
- MCP
- Model Context Protocol, an open protocol for connecting AI applications to tools, data sources, and prompts through defined interfaces.
- Evaluation
- A repeatable method for measuring a system against defined quality, safety, or task criteria.
- Tool Calling
- A model interface pattern where the model requests a structured operation from an external tool.
- API Layer
- The server-side interface that exposes defined data and actions to clients or integrations.
- 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.
- Anthropic adds code execution, MCP connector, Files API, and caching
Anthropic announced four API capabilities for agents: a code-execution tool, an MCP connector, a Files API, and extended prompt caching.
- Anthropic adds remote MCP support to Claude Code
Anthropic announced remote Model Context Protocol support in Claude Code. The company said developers could connect Claude Code to remote MCP servers by adding a vendor URL instead of operating local servers.

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.