Agentic apps registration approval and identity badges

Verified 2026-09-05 · 56 claims · sources tier 2 · 5 disputed

Also known as A2A, ACS, AGNTCY, Agent Central, Agentic App, Cisco Official, Federated through Registry, OASF.

Cisco defines an agentic app as a service exposing structured capabilities that an artificial intelligence agent can invoke, which is not tied to a specific protocol and is analogous to a Model Context Protocol server when that standard is used 11. Provisioning and governance for these apps take place within Control Hub to regulate authentication, user access, and tool availability 25​40.

Provisioning and governance in Control Hub

Agentic app provisioning is located in Control Hub under Apps and then Agentic Apps, displaying both public applications and private apps onboarded by users within the organisation 25. Documentation reflects inconsistent navigation terminology, as 2 Cisco documents give different Control Hub paths: 1 specifies Apps and then Agentic Apps, while the other specifies Management, then Apps, then Agentic Servers 26. The Control Hub provisioning guide is intended for administrators only, establishing governance, authentication, and access controls over available agentic capabilities 40. The provisioning documentation also states that users bear full responsibility for understanding, complying with, and managing all aspects of the liability framework regarding Model Context Protocol application usage, though it never defines what that framework is or where it is published 37.

On an agentic app's general tab in Control Hub, administrators can allow or block the app for all organisation users 34. Sources disagree on the availability of group-level access controls: Cisco's admin approval guide documents an enablement audience of either the entire organisation or selected user groups resolved from the identity provider at request time 8, whereas Cisco's Model Context Protocol documentation for AI Agent Studio states that group-level access control is not supported and tool access is managed exclusively at the organisation level 48. Sources disagree further regarding per-tool assignments, as the admin approval guide outlines a per-tool assignment mode where each tool can be enabled for a different set of groups, overriding the server-level audience 39.

Cisco recommends evaluating tool risk based on the tool's name, description, input schema, and stated side effects, advising administrators to restrict operations involving deletion, financial transactions, sending email, and modifying user data to specific groups 35. However, tools carry no formal risk class field, meaning an administrator must evaluate the risk level using their own judgement 36. Additionally, an authorise automatic server data updates setting allows changes to a server's name, description, transport type, other metadata, or its address within the same domain to take effect without re-authorisation, while presenting an indicator to the administrator that changes occurred 23.

Capabilities and tool management

Control Hub presents 3 capability tabs for a Model Context Protocol server, covering tools, resources, and prompts 49. Enabling a tool in Control Hub makes it visible, discoverable, and usable by organisation users, while disabling it removes visibility and access 51. Prior to enabling a tool, an administrator can review its input schema, output schema, and annotations directly in Control Hub 7. Cisco states that similar functionality exists on the resources and prompts pages, but WarmTransfer's reading of the sources is that it does not itemise what similar covers 43.

Once created in AI Agent Studio, Model Context Protocol actions are read-only, requiring a new action to be created if settings must be changed 38. WarmTransfer's reading of the sources is that the developer portal does not publish a specific number for the maximum action limit per agent, stating only that it is defined by the organisation's configuration 6.

Schema signature monitoring and drift

Cisco computes a schema signature for each tool and monitors it continuously, flagging any tool whose live signature differs from the previously approved version 44. Runtime behaviour upon signature drift is governed by the allow signature change setting 45​46:

• If the allow signature change setting is off, drifting tools are filtered out of list functions results and omitted from client delivery 45.
• If the allow signature change setting is on, drifting tools continue to be delivered but carry an unsecure flag set to true so agent interfaces and clients can display warnings 46.

Regardless of whether the signature change setting permits continuous runtime access, formal approval of a new tool schema always requires explicit administrative re-authorisation 41.

Authentication methods

The authentication details requested in Control Hub correspond to the configuration specified in the developer portal 22. Cisco official servers have their authentication configured automatically, requiring administrative edits only when overriding defaults 24. Control Hub provides 5 authentication methods: OAuth2 client credentials, OAuth2 authorization code, user token, application key, and custom headers 32.

Input requirements vary across these methods:

• For OAuth2 client credentials the token endpoint authentication method may be client secret basic or client secret post 50.
• OAuth2 authorization code requires a registration endpoint and authorization endpoint in addition to the client credentials fields, and permits its token endpoint authentication method to be client secret basic, client secret post, or none 21​50.
• Application key requires only the key itself 21.
• User token requires no administrator input 21.
• Custom headers support up to 5 key and value pairs across all authentication types 27.

Nomenclature differs between sources: the admin approval guide outlines 4 authentication types under machine names such as organisation-scoped token and user-scoped token, which do not align with the 5 method names in Control Hub 9. Furthermore, the admin approval guide attributes authentication configuration to Agent Central, a entity name that appears neither on the Control Hub provisioning page nor in other cited documentation 10.

In AI Agent Studio, tools using OAuth2 authorization code are currently omitted from the available actions list because the interface displays only tools with supported authentication methods 47.

App Hub identity badges and registration

Submitting a registered agentic app to the Webex App Hub is optional and initiated from the developer portal; unsubmitted apps remain functional within the developer's organisation 12. Generic App Hub submissions involve a 2-phase review covering marketing materials followed by functionality and usability testing 19. WarmTransfer's reading of the sources is that the main App Hub submission page focuses broadly on embedded apps, integrations, and bots without referencing agentic apps or the Model Context Protocol, while a separate dedicated page addresses agentic app submission and badging 20.

Cisco establishes 5 identity badges for agentic apps on the App Hub, ranked in descending order of assurance: Cisco official, Cisco onboarded, partner onboarded, developer onboarded, and federated through registry 33.

Across the Webex App Hub, 39 agentic apps are reported in total 16. In the initial batch of 32 returned listings, 4 have the Cisco official badge, 6 carry Cisco onboarded, and 22 carry federated through registry, and none carries the partner onboarded or developer onboarded badge 14. WarmTransfer's reading of the sources is that marketplace popularity sorting lacks differentiated data, as sampled listings show an empty published date and identical popularity values 18. The App Hub includes a blanket disclaimer stating third-party apps are provided by unaffiliated developers under their own terms, disclaiming Cisco responsibility or liability 15.

Federated registry servers

Federated through external registry servers are third-party products imported without Cisco ownership or manual review, and functionality issues must be directed to the third-party developer 28​29. WarmTransfer's reading of the sources is that external federation is active and deployed rather than planned work, as the documentation features no preview qualifiers and all listed servers appear in the marketplace 31.

The documentation lists 22 federated servers spanning search, scraping, document signature, product analytics, automation, calendar, workspace, payment, developer tooling, database, and network monitoring 52. All 22 federated servers use OAuth2 authorization code authentication 13. Because AI Agent Studio excludes tools configured with OAuth2 authorization code 47, WarmTransfer's reading of the sources is that federated servers cannot currently be added as actions in AI Agent Studio 30. Additionally, naming discrepancies exist between platforms: 4 federated servers are spelled differently in the developer portal listing compared to the App Hub listing despite describing identical products and addresses 17.

Agent-to-agent protocol status

Sources disagree regarding the availability of agent-to-agent registration: Cisco states agent-to-agent registration is in beta and requires enrolling in a beta programme 5, whereas the module selector on the Model Context Protocol registration form marks agent-to-agent as coming soon 42. The agent-to-agent registration page is located at a path ending in onboard your agent, differing from names cross-referenced in the AI Agent Studio guide 4.

Technically, agent-to-agent registration requires hosting an agent card conforming to version 1 of the agent-to-agent protocol at a well-known path on the server's base address, which the portal fetches and validates 1. Authentication and transport configurations are auto-discovered from this agent card and do not require manual administrative entry during registration 2. While the App Hub provides an agent-to-agent filter, all 32 sampled listings carry the Model Context Protocol badge, with none carrying the agent-to-agent badge 3.

See also

• Model Context Protocol actions in Webex contact centre AI
• Tool schema signature monitoring and drift handling
• The Webex App Hub as a marketplace
• Webex AI Agent Studio scripted and autonomous agents
• Webex published MCP servers for external AI clients

Applicability

Across the claims in this article, the evidence covers Cisco Webex across cloud deployments, verified as of 2026-09-05.

What remains uncertain

Specific field-level differences between the tools capability tab and the resources and prompts capability tabs are not covered by any claim in this article. The identity and documentation location of the Agent Central platform mentioned in the admin approval guide are not covered by any claim in this article. The precise maximum action limit enforced per agent in AI Agent Studio is not covered by any claim in this article. The exact criteria and review requirements for obtaining the partner onboarded and developer onboarded App Hub badges are not covered by any claim in this article. The complete contents and publisher of the liability framework referenced in the Control Hub provisioning guide are not covered by any claim in this article.

WarmTransfer. (2026, September 5). Agentic apps registration approval and identity badges. WarmTransfer. https://warmtransfer.net/knowledge/webex-agentic-apps-governance

@misc{warmtransfer-webex-agentic-apps-governance,
  title  = {Agentic apps registration approval and identity badges},
  author = {{WarmTransfer}},
  year   = {2026},
  url    = {https://warmtransfer.net/knowledge/webex-agentic-apps-governance},
  note   = {Verified 2026-09-05}
}