Skip to main content
This page lists the workspace secrets you need to add before using Agent Builder. Add these in your LangSmith workspace settings under Secrets. Keep values scoped to your workspace and avoid placing credentials in prompts or code.

​
How to add workspace secrets

In the LangSmith UI, ensure that you have an LLM API key set as a workspace secret (either Anthropic or OpenAI API key).
  1. Navigate to Settings and then move to the Secrets tab.
  2. Select Add secret and enter either OPENAI_API_KEY or ANTHROPIC_API_KEY as the name, and your API key as the value.
  3. Select Save secret.
When adding workspace secrets in the LangSmith UI, make sure the secret keys match the environment variable names expected by your model provider.

​
Required model key

For Agent Builder to make API calls to LLMs, you need to set an OpenAI or Anthropic API key as a workspace secret. The agent graphs load this key from workspace secrets for inference.
Agent Builder supports custom models per agent. See Custom models for more information.

​
Agent Builder specific secrets

Secrets prefixed with AGENT_BUILDER_ are prioritized over workspace secrets within Agent Builder. This way, you can better track the usage of Agent Builder vs other parts of LangSmith which use the same secrets. If you have both OPENAI_API_KEY and AGENT_BUILDER_OPENAI_API_KEY, the AGENT_BUILDER_OPENAI_API_KEY secret will be used.

​
Optional tool keys

Add keys for any tools you enable. These are read from workspace secrets at runtime.
  • EXA_API_KEY: Required for Exa search tools (general web and LinkedIn profile search).
  • TAVILY_API_KEY: Required for Tavily web search.
  • TWITTER_API_KEY and TWITTER_API_KEY_SECRET: Required for Twitter/X read operations (app‑only bearer). Posting/media upload is not enabled.

​
MCP server configuration

Agent Builder can pull tools from one or more remote Model Context Protocol (MCP) servers. Configure MCP servers and headers in your workspace settings. Agent Builder automatically discovers tools and applies the configured headers when calling them. For more details on using the remote MCP servers, refer to the the MCP Framework page.
Edit the source of this page on GitHub.
Connect these docs programmatically to Claude, VSCode, and more via MCP for real-time answers.