Model provider
Anthropic
Agents whose Runtime is claude-mcp call the Anthropic API and may use their allowed MCP servers as tools.
The claude-mcp Runtime offers the model the tools of the MCP servers the Agent may use. Every tool call goes through the Tool broker, which re-checks the allow-list, validates the input and records an Audit event. An Agent run takes at most 8 tool steps.
Like openai, the Runtime reports itself unsupported without a key, and an Invoke then fails cleanly.
More integrations
All integrations →- 01
Yellow Pages
Helmsman Registers its Agents in the Fleetworks service and agent catalog and listens for the catalog’s signed webhooks.
- 02
Zitadel
People sign in through Zitadel, the shared Fleetworks identity provider, with the authorization code flow and PKCE.
- 03
OpenAI
Agents whose Runtime is openai call the OpenAI API when invoked.