Claude Managed Agents marketplace: which surface fits?

If you search for a Claude Managed Agents marketplace, you can land on two very different surfaces. Anthropic's Claude Marketplace is a partner marketplace for connectors, plugins, agents and products, and service partners. Coduo's Managed Agents is a catalog for finding Anthropic Managed Agent packages that you can inspect before applying them in your own workspace. The names are close enough to cause a costly mix-up.

The distinction matters because the two surfaces answer different questions. The official Marketplace helps a team find a connector, buy partner software or find help with a Claude rollout. A package catalog helps a person decide whether a folder of agents, environments, skills, memory stores or deployments is suitable for their own work. Neither surface should be described as the other. Start with the Managed Agents catalog when your task is to inspect and apply a package, and use the catalog's examples guide when you want to compare package shapes.

What does the Claude Marketplace contain?

The official Claude Marketplace contains connectors and plugins, agents and products from partners, and service partners. Anthropic describes connectors and plugins as a way to connect apps, data and tools to Claude. Its agents and products section is for Claude-powered software from partners. Its service-partner section is for certified experts that help organizations implement and scale Claude.

That makes Claude Marketplace the right baseline for commercial partner offerings and integrations. If you need to connect a workplace tool, evaluate partner software or find implementation help, begin there. The page is also the primary source for what Anthropic currently calls the Marketplace, so its categories and wording should win over a third-party directory's label.

Claude Managed Agents is related to the platform, but it is not a fourth Marketplace category. Anthropic's Managed Agents documentation describes a managed agent as a configurable harness for long-running or asynchronous work. It separates an agent, an environment, a session and events. Those are the platform concepts a package may define. Claude Marketplace is the partner surface where Anthropic groups products and services. Managed Agents is the Coduo catalog where a visitor can inspect a package record before using those platform concepts in a workspace.

The last row is a recommendation based on the two surfaces' documented jobs, not a claim that one marketplace is better. A reader looking for a partner product should not be sent to a package catalog. A reader looking for a downloadable package should not have to infer its resource list from a commercial product page.

Where can I find a downloadable Managed Agent package?

Use the Managed Agents catalog when you want a package record that describes an outcome, requirements and verification evidence before you apply anything. The catalog contract says one listing describes one versioned offering. It should give a visitor enough information to understand the outcome, inspect requirements, download an artifact when one exists and reproduce the stated verification level.

The catalog home is the fastest way to narrow the task. Search by a phrase such as SEO, planning or research, then use category and tag controls to reduce the list. The public catalog routes and fixture adapter support query, category, outcome and tag filters. A listing card puts the outcome and cadence before the lower-level runtime details, so you can reject a poor fit without opening every package.

The item page is where discovery becomes an install decision. Read the package README, native resource paths, requirements, example output, schedule and testing section. The page can show an install command for running once, a separate command for applying a deployment schedule, and a note about what an interactive session must attach. These are package-specific details. They are not promises that a package has run successfully in your workspace.

Download is conditional. The detail code only creates a download URL when the listing has a license, a package version, a valid SHA-256 identity, and either a stored package or safe artifact metadata. That check protects the boundary between a listing that is readable and an archive that the site is willing to serve. If one of those fields is unresolved, treat the package as a review candidate rather than silently filling in the gap.

Managed Agents remains a catalog, not a runner. It does not execute ant apply, hold an Anthropic credential or schedule your work. The package archive moves to your workspace, where you inspect it and decide what to apply. For the platform's resource model, the Anthropic documentation is the baseline. For package discovery, provenance and evidence, the catalog is the useful next step.

What should I inspect before I apply a package?

Inspect the package's intended outcome, resources, access, schedule, identity and evidence before you apply it. The following order keeps the attractive part of an example, its output, from hiding the operational details.

1. Outcome and expected output. Read the one-line outcome, jobs and audience. Then check the example output and required inputs. A package that promises a weekly report may still need a specific site URL, data source or user profile. If the inputs are vague, the output is not a guarantee.
2. Native resources. Check every required and optional path. The catalog contract permits agents, environments, skills, memory_stores, deployments and vaults. A vault file can describe metadata without carrying a secret. A workspace lockfile or token should not be part of the archive.
3. Runtime access. Read network policy, MCP connections, CLI packages, skills, memory policy and permissions as one group. An empty list means that no requirement was observed in the record. It does not prove that a runtime was tested without that capability.
4. Cadence and cost notes. A package may run on demand, or it may include a deployment. A deployment is a scheduled creator of sessions. The detail page warns that applying one can start billed runs, so look for the run-once path before you choose the command that applies the schedule.
5. Provenance and artifact identity. Look for source repository, source commit, license, package version, artifact URL and SHA-256. These fields tell you what you are downloading and whether the archive can be matched to the catalog record. Missing provenance is a review gap, not a detail to guess.
6. Verification level and limitations. Keep imported, converted, deployable, functional and verified separate. Static shape checks show that files and references pass a gate. They do not prove that an authenticated apply succeeded, that a representative task ran, or that memory and scheduling worked as expected.

The last distinction is the one most generic marketplace lists skip. A polished README and a clean archive can still stop at static validation. A live test record can still have a narrower scope than your workspace. Read the date, CLI or platform context, fixture and limitations of the evidence before treating a badge as a fit decision.

Which marketplace or catalog fits my install task?

Choose Claude Marketplace for connectors, partner agents and products, or service-partner help. Choose Managed Agents when the task is to find a native package, inspect its files and requirements, and decide what to apply in your own Claude workspace. Use the official Marketplace as the authority for Anthropic's partner offerings and the catalog record as the authority for the package you are evaluating.

For a package install, make the next step concrete:

1. Open Managed Agents search and filter by the outcome you need.
2. Open a listing and read its example output, requirements, schedule, permissions, provenance and testing evidence.
3. Download only when the listing exposes a complete artifact identity and license.
4. Move the archive into your own workspace, authenticate ant, run the documented dry run, and read the plan before applying it.

That workflow leaves the commercial partner question and the package question in the right places. If you want more examples before choosing, compare package examples by outcome. If you already know the package you want, open its install section in Managed Agents and treat the catalog as the inspection stop before your workspace becomes the execution stop.