Claude Code

Claude Code can call Openhandle directly through the MCP server. Add it once, and every profile, post, comment, and search operation becomes a tool the agent can use.

1. Install the plugin

Run these commands in Claude Code:

/plugin marketplace add openhandlehq/openhandle-claude
/plugin install openhandle@openhandle

The plugin adds the MCP server and two skills. One skill teaches Claude to use the tools. The other adds Openhandle to your application code.

2. Sign in

Type /mcp, pick openhandle, and approve in the browser. Pick Test on that screen. Test runs every tool against synthetic data and is never charged. For real accounts, clear the authentication in /mcp, approve again, and pick Live.

Use an API key instead

Scripts and headless setups can skip the browser. Create a Test key in the dashboard and add the server with it:

claude mcp add --transport http openhandle https://api.openhandle.dev/mcp \
  --header "Authorization: Bearer $OPENHANDLE_TEST_KEY"

The key decides Test or Live. Do not add an environment flag anywhere.

3. Ask

Find a public Instagram profile in the test data and show me its follower count and last three posts.

Claude Code reads the tool list, calls find_test_data to pick a synthetic account, and then calls the profile and posts tools. Every result carries capturedAt, source, and the charge, so the agent can tell you what it cost.

Notes

• Ask about your account in plain words, for example "What did I spend on Instagram this week?" The account tools read your balance, usage, request history, and API keys.
• One prompt sets up a whole project: ask the agent to read openhandle.dev/SKILL.md and follow it.
• Every docs page is markdown. Add .md to any URL.
• The MCP endpoint and tool list are described in Test with MCP.