The searches are the raw material; the agent runs are what make it more than a query tool — one plain-English question produces a markdown report with signals and evidence links, and a schedule reruns it weekly. Oversized responses come back truncated by design, so plan on fetching the raw posts one platform at a time when you need the underlying text.
The MCP server for Rolli IQ. It runs three kinds of search — keyword or hashtag across platforms, a user profile, and an expert recommendation from a plain-language brief — and layers an agent on top that plans its own search strategy and writes a markdown report, on demand or on a schedule.
- keyword_search, get_keyword_search and list_keyword_searches — create a cross-platform keyword or hashtag search, poll it to completion, and revisit past ones by status
- user_search, get_user_search and list_user_searches — a profile search returning metrics and content analysis
- get_keyword_search_posts and get_user_search_posts — the raw posts behind a search when the summary response was truncated
- get_topic_tree — how topics and subtopics are distributed across a keyword search's results
- expert_search, get_expert_search and list_expert_searches — find people matching a natural-language query, returned with AI-generated match summaries
- start_agent_run, get_agent_run, cancel_agent_run, rerun_agent_run, update_agent_run, delete_agent_run and list_agent_runs — submit a plain-English question, watch the phase and percentage while it works, and read the markdown report with its signals and evidence links
- create_agent_schedule, get_agent_schedule, update_agent_schedule, pause_agent_schedule, resume_agent_schedule, delete_agent_schedule and list_agent_schedule_runs — recurring reports on a daily, weekly, monthly or one-time cadence, with email and Slack notifications
- get_integration_setup, update_integration_setup and get_usage — the webhook that fires when a search completes, and search counts with the remaining IQ credit balance
A Rolli account with API access and its token in ROLLI_API_TOKEN. Clients run npx -y @rolli/mcp over stdio. ROLLI_USER_ID is optional and defaults to rolli-mcp. Searches poll to completion inside the tool call, so ROLLI_POLL_INTERVAL_MS and ROLLI_MAX_POLL_MS control how long a call is willing to wait — the default ceiling is ten minutes.
One command plus a key — npx -y @rolli/mcp, then supply credentials