MCP server
Feedpik speaks the Model Context Protocol, so Claude, Cursor, and any other MCP client can read your ranked candidates and tune your ranking as native tools — no shell required. If your agent does have a shell, the agent skill and CLI cover the same ground.
Install
Add this to your MCP client’s config — claude_desktop_config.json for Claude Desktop, or the equivalent for Cursor and Claude Code:
{
"mcpServers": {
"feedpik": {
"command": "npx",
"args": ["-y", "-p", "feedpik", "feedpik-mcp"]
}
}
}Point it at a different server with FEEDPIK_API_URL when you need to.
Signing in
The server has no key of its own. Ask your assistant to connect to Feedpik: it calls feedpik_login_start, which returns a short code and a URL. Open the URL, review exactly which permissions are being requested, and approve. The key is saved to ~/.feedpik/config.json and shared with the CLI, so you only do this once per machine.
Read access is the default. Write scopes — changing your ranking, adding subscriptions — are only ever granted if they are spelled out on that approval screen. You can revoke any key under Algorithm & Agents → Agents & Access.
Tools
feedpik_list_candidatesRanked candidates for the active persona, each with its reasonfeedpik_record_feedbackRecord read / saved / liked / skipped / negative feedbackfeedpik_get_policyThe active ranking policy and its version historyfeedpik_update_policySave a modified ranking config as a new, rollbackable versionfeedpik_rollback_policyRestore a previous policy versionfeedpik_list_personasPersonas — each is a whole reading spacefeedpik_switch_personaSwitch which persona every other tool acts onfeedpik_list_feedsSubscriptions of the active personafeedpik_add_feedSubscribe to an RSS/Atom feedfeedpik_remove_feedUnsubscribe from the active persona onlyfeedpik_discover_sourcesSearch the source catalog or browse a categoryfeedpik_list_packsCommunity and official source packsfeedpik_install_packAdd every source in a pack to the active personafeedpik_whoamiCurrent token, scopes, and account statusfeedpik_login_startBegin authorizing — returns a code for the user to approvefeedpik_login_completeCheck whether the user approved the login