Interface: OpenAIAgentsPluginOptions
openaiAgents.OpenAIAgentsPluginOptions
Properties
hostedToolCredentials
• Optional hostedToolCredentials: HostedToolCredentialsResolver
Resolves credentials for the hosted tools an Agent declares — a hosted MCP
tool's authorization and headers, a shell or code interpreter tool's
domain secrets — so Workflow code can omit them and keep them out of
Workflow history. See HostedToolCredentialsResolver.
Worker-side hosted tool credentials are experimental and may change without notice.
interceptorOptions
• Optional interceptorOptions: OpenAIAgentsPluginInterceptorOptions
Trace interceptor options. Propagated to the Workflow via the
__openai_agents_config header and applied to Activities directly.
mcpServerProviders
• Optional mcpServerProviders: MCPServerProvider[]
MCP server providers whose Activities will be auto-registered.
modelParams
• Optional modelParams: SerializableModelActivityOptions
Default Model Activity options (timeouts, retry, Task Queue, etc.).
Propagated to the Workflow via the __openai_agents_config header.
modelProvider
• modelProvider: ModelProvider
Model provider used to resolve model names to Model instances.
resolvableWorkerEnvVars
• Optional resolvableWorkerEnvVars: readonly string[]
Allowlist of Worker environment variable names this Worker will read on
behalf of a workerEnvValue in a sandbox Manifest. Names must match
exactly, except for the entry '*', which allows every name.
An allowlisted variable that is unset or empty reads as the empty string. A name outside the allowlist fails the Activity non-retryably.
Default
[] — no name is readable.
@experimental Worker environment variable references are experimental and may change without notice.
sandboxClientProviders
• Optional sandboxClientProviders: SandboxClientProvider[]
Sandbox client providers whose Activities will be auto-registered.
Reference them from Workflows via temporalSandboxClient(name).
Sandbox support is experimental and may change without notice.