Connect your agent to Hedra tools.

Explore models and generate synthetic data for perception research, directly from your AI assistant. Inspect inputs, estimate costs, and retrieve results through Hedra.

ClaudeClaude CodeCodexCursorChatGPTHermes

Connect with MCP or the CLI.

Connect

Connect with MCP or the CLI

MCPCLI

1. 1
   Add Hedra to ClaudeOpen Claude with the Hedra connector name and URL already filled in.Add to Claude
2. 2
   Review and connectConfirm the connector, then click Connect to start the Hedra sign-in.
3. 3
   Sign in and askFinish the Hedra sign-in, then try this in the chat. Pay as you go or use your Hedra subscription.

   Use Hedra to generate one synthetic image for a scene-understanding experiment: a workbench with three distinct tools, a partially obscured box, and an empty shelf. Choose a suitable image model, check its inputs and estimated cost, then generate the image. Return the result with the model and parameters so I can reproduce the setup.Copy

1. 1
   Add the serverRun this in any project. One command, no config file to edit.

   claude mcp add --transport http hedra https://mcp.hedra.comCopy
2. 2
   Sign inRun /mcp and complete the Hedra sign-in in the browser window it opens.
3. 3
   Inspect available modelsStart by checking the models and tools available through your connection.

   Use Hedra to generate one synthetic image for a scene-understanding experiment: a workbench with three distinct tools, a partially obscured box, and an empty shelf. Choose a suitable image model, check its inputs and estimated cost, then generate the image. Return the result with the model and parameters so I can reproduce the setup.Copy

1. 1
   Add the serverRun this once — Codex stores it in your config.

   codex mcp add hedra --url https://mcp.hedra.comCopy
2. 2
   Approve the sign-inStart Codex and approve the Hedra sign-in when it prompts you.
3. 3
   Inspect available modelsTry this in the conversation to inspect model inputs before running a job.

   Use Hedra to generate one synthetic image for a scene-understanding experiment: a workbench with three distinct tools, a partially obscured box, and an empty shelf. Choose a suitable image model, check its inputs and estimated cost, then generate the image. Return the result with the model and parameters so I can reproduce the setup.Copy

1. 1
   Copy the configCursor reads MCP servers from a JSON file.

   { "mcpServers": { "hedra": { "url": "https://mcp.hedra.com" } } }Copy
2. 2
   Add it to CursorOne click installs it, or paste the block above into ~/.cursor/mcp.json yourself.Add to Cursor
3. 3
   Reload and sign inReload Cursor, then sign in to Hedra from the MCP panel and try this in chat.

   Use Hedra to generate one synthetic image for a scene-understanding experiment: a workbench with three distinct tools, a partially obscured box, and an empty shelf. Choose a suitable image model, check its inputs and estimated cost, then generate the image. Return the result with the model and parameters so I can reproduce the setup.Copy

1. 1
   Copy the Hedra URLClick copy — you'll paste it in the next step.

   https://mcp.hedra.comCopy
2. 2
   Settings → AppsAdd a custom app and paste the URL. Plans that gate this need developer mode turned on first.Open ChatGPT apps
3. 3
   Authorize and askAuthorize Hedra, then try this in any chat.

   Use Hedra to generate one synthetic image for a scene-understanding experiment: a workbench with three distinct tools, a partially obscured box, and an empty shelf. Choose a suitable image model, check its inputs and estimated cost, then generate the image. Return the result with the model and parameters so I can reproduce the setup.Copy

1. 1
   Add the serverHermes reads MCP servers from ~/.hermes/config.yaml.

   mcp_servers: hedra: url: "https://mcp.hedra.com" auth: oauthCopy
2. 2
   AuthorizeHermes opens an authorize URL and waits for the OAuth callback, then caches the token.
3. 3
   Inspect available modelsHedra's tools show up alongside the rest of your Hermes toolset. Try this one first.

   Use Hedra to generate one synthetic image for a scene-understanding experiment: a workbench with three distinct tools, a partially obscured box, and an empty shelf. Choose a suitable image model, check its inputs and estimated cost, then generate the image. Return the result with the model and parameters so I can reproduce the setup.Copy