Rolli MCP
工具
工具 | 描述 |
| 列出所有关键词搜索,并按状态过滤 |
| 创建关键词/标签搜索并在完成后返回结果 |
| 获取关键词搜索结果(状态、分析、帖子) |
| 列出所有用户搜索,并按状态过滤 |
| 创建用户资料搜索并在完成后返回结果 |
| 获取用户搜索结果(资料、指标、内容分析) |
| 获取关键词搜索的对话主题树 |
| 获取关键词搜索的原始帖子 |
| 获取用户搜索的原始帖子 |
| 获取当前集成设置(Webhook URL、名称) |
| 设置用于搜索完成通知的 Webhook URL |
| 获取 API 使用数据和各用户的明细 |
Related MCP server: Grok X Insights MCP Server
设置
你需要一个拥有 API 访问权限的 Rolli 账户。请从 rolli.ai 获取你的 API 令牌。
Claude Desktop / Claude Code / VS Code / Cursor / Windsurf
添加到你的 MCP 配置中(claude_desktop_config.json、.vscode/mcp.json 或等效文件):
{
"mcpServers": {
"rolli": {
"command": "npx",
"args": ["-y", "@rolli/mcp"],
"env": {
"ROLLI_API_TOKEN": "your_token"
}
}
}
}ROLLI_USER_ID 是可选的,默认为 "rolli-mcp"。设置此项可覆盖为你自己的用户 ID。
可选环境变量
变量 | 默认值 | 描述 |
|
| 随 API 请求发送的用户 ID |
|
| 等待搜索结果时的轮询间隔(毫秒) |
|
| 搜索超时前的最长等待时间(毫秒)(默认:10 分钟) |
Smithery
此服务器包含一个用于通过 Smithery 部署的 smithery.yaml。它会在设置过程中提示你输入 API 令牌。
使用示例
在社交媒体上搜索关键词:
"Search for posts about 'artificial intelligence' on Twitter and Reddit from the last week"
分析用户资料:
"Look up @elonmusk on Twitter and analyze their recent posts"
获取主题细分:
"Show me the topic tree for my keyword search #123"
检查 API 使用情况:
"How many searches have I used this month?"
设置 Webhook:
"Set my webhook URL to https://myapp.com/rolli-callback"
开发
npm ci
npm run lint # ESLint
npm run build # TypeScript
npm test # VitestCI 在每次推送到 master 分支和提交 PR 时运行。在 GitHub 上发布的版本会自动发布到 npm(需要 NPM_TOKEN 仓库密钥)。
许可证
MIT
Available Tools
30 toolscancel_agent_runA
Cancel a pending or running Rolli Agent run. Marks the run as failed.
Inputs: id (required).
create_agent_scheduleB
Create a recurring Rolli Agent report that runs on a cron-like cadence (daily, weekly, monthly, or one-time) with optional email and Slack notifications.
Inputs: name (required), question (required), starts_at (required), recurrence_time (required), +16 more.
delete_agent_runA
Soft-delete a Rolli Agent run. The record remains in the database for audit purposes but is hidden from listings. Only works on completed or failed runs — cancel the run first if it is still running.
Inputs: id (required).
delete_agent_scheduleA
Permanently delete a Rolli Agent schedule and its execution history. This is irreversible. Use pause_agent_schedule to temporarily stop a schedule without losing history.
Inputs: id (required).
expert_searchA
Find experts matching a natural-language query using Rolli's AI-driven recommendation engine. Polls until the search is complete and returns the full list of recommended experts (name, professional title, location, contact info, expertise keywords, and an AI-generated summary explaining why each expert matches). A status of results_pending means the search succeeded but results are still being written; fetch them with get_expert_search.
Inputs: query (required).
get_agent_runA
Get a Rolli Agent run by ID. While running, returns progress (percentage, phase, current tool). Once complete, returns the full markdown report, signals, evidence links, and usage metrics.
Inputs: id (required).
get_agent_scheduleA
Get a Rolli Agent schedule's full configuration plus its 10 most recent execution records.
Inputs: id (required).
get_expert_searchA
Get results for an expert search by ID. Returns search status and, once complete, the array of recommended experts with their profiles and AI-generated match summaries. results_pending: true means the search succeeded but the backend is still writing results; retry in a few seconds.
Inputs: id (required).
get_integration_setupA
Get the current integration settings (webhook return URL and integration name).
Inputs: no named parameters.
get_keyword_searchA
Get results for a keyword search by ID. Returns search status, analytics summary, and posts. Oversized responses are truncated (truncated: true) — read the complete post list in pages with get_keyword_search_posts (offset, limit, fields=lean, one platform at a time). results_pending: true means the search succeeded but the backend is still writing results; retry in a few seconds.
Inputs: id (required).
get_keyword_search_postsA
Get raw posts from a keyword search. The first line says how many of the search's posts this response shows and the offset of the next page; total, returned, offset and platform_counts (posts per platform across the whole search) are in the JSON. Read a large search in pages with offset and limit, one platform at a time, and fields=lean when the text matters more than media and profile URLs. A response over the size cap is marked truncated: true.
Inputs: search_id (required), limit, fields, offset, +1 more.
get_topic_treeB
Get the conversation topic tree for a keyword search. Shows how topics and subtopics are distributed across the search results.
Inputs: search_id (required), platform.
get_usageA
Get API usage data. Returns search counts, per-user breakdowns, and the current IQ credit balance (credits.credits_remaining, credits.credits_limit, credits.period_reset_date — the latter two are null for one-time credit grants). Optionally filter by month.
Inputs: month.
get_user_searchA
Get results for a user search by ID. Returns profile info, metrics, and content analysis. Oversized responses are truncated (truncated: true) — read the complete post list in pages with get_user_search_posts (offset, limit, fields=lean). results_pending: true means the search succeeded but the backend is still writing results; retry in a few seconds.
Inputs: id (required).
get_user_search_postsA
Get raw posts from a user search. The first line says how many of the profile's posts this response shows and the offset of the next page; total, returned and offset are in the JSON. Read a large search in pages with offset and limit, and fields=lean when the text matters more than media and profile URLs. A response over the size cap is marked truncated: true.
Inputs: search_id (required), limit, fields, offset.
keyword_searchA
Create a keyword/hashtag search across social media platforms (X, Reddit, YouTube, Facebook, Instagram, Threads, Bluesky, and more). Polls until the search is complete and returns the full results. Oversized responses are truncated (truncated: true, largest post arrays shortened) — read the complete post list in pages with get_keyword_search_posts (offset, limit, fields=lean, one platform at a time). A status of results_pending means the search succeeded but results are still being written; fetch them with get_keyword_search.
Inputs: query (required), sort_by, end_date, max_post, +4 more.
list_agent_runsA
List Rolli Agent runs (AI-driven social intelligence investigations). Returns a paginated summary list. Discarded runs are excluded.
Inputs: limit, offset, status, user_id.
list_agent_schedule_runsA
List the full execution history of a Rolli Agent schedule, including a preview of each linked agent run.
Inputs: id (required), limit, offset.
list_agent_schedulesB
List Rolli Agent scheduled reports. Schedules execute on a daily/weekly/monthly/once cadence and produce a new agent run each time.
Inputs: limit, offset.
list_expert_searchesC
List all expert searches. Returns a paginated list filtered by status.
Inputs: page, show.
list_keyword_searchesB
List all keyword searches. Returns a paginated list filtered by status.
Inputs: page, show.
list_user_searchesC
List all user searches. Returns a paginated list filtered by status.
Inputs: page, show.
pause_agent_scheduleA
Pause a Rolli Agent schedule so it stops executing. The schedule and its history are preserved — resume it later with resume_agent_schedule.
Inputs: id (required).
rerun_agent_runA
Rerun a failed, waiting, or stuck Rolli Agent run. Resets the run to pending and re-enqueues it. Running runs are force-cancelled first. Returns 422 if the run is already completed. Fails immediately if the IQ credit balance is exhausted (a rerun needs at least 1 credit).
Inputs: id (required).
resume_agent_scheduleA
Resume a paused Rolli Agent schedule. next_run_at is recalculated based on the recurrence parameters.
Inputs: id (required).
start_agent_runA
Start a Rolli Agent investigation. Submit a plain-English question; the agent plans a search strategy, queries Rolli IQ across platforms, analyzes posts, and produces a markdown intelligence report with signals, evidence links, and a confidence rating. Fails immediately (before starting) if the IQ credit balance is insufficient. Polls inline for up to 2 minutes. If the run finishes in that window, returns the full results; otherwise returns the run ID and a note to fetch the results with get_agent_run (typical run takes 3–5 minutes).
Inputs: question (required), mode, max_post, metadata, +4 more.
update_agent_runB
Update a Rolli Agent run's metadata (rename or pin). At least one of pinned or label is required.
Inputs: id (required), label, pinned.
update_agent_scheduleB
Partially update a Rolli Agent schedule. Unspecified fields are left unchanged. next_run_at is recalculated if any recurrence parameter changes.
Inputs: id (required), mode, name, ends_at, +14 more.
update_integration_setupC
Update integration configuration. Set the webhook URL that receives notifications when a search completes.
Inputs: return_url (required).
user_searchA
Create a user profile search on a social media platform. Polls until the search is complete and returns the full results. Oversized responses are truncated (truncated: true) — read the complete post list in pages with get_user_search_posts (offset, limit, fields=lean). A status of results_pending means the search succeeded but results are still being written; fetch them with get_user_search.
Inputs: query (required), platform (required), end_date, enable_ai, +1 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 30 tools
The three search creators (keyword_search, user_search, expert_search) target genuinely distinct resources, and the get_/list_/get_*_posts variants are separated by clear descriptions about paging and polling. The only friction is the large family of similarly named search readers (get_keyword_search vs. get_keyword_search_posts vs. list_keyword_searches), which an agent must read carefully but can still distinguish.
Nearly everything follows snake_case verb_noun (get_agent_run, create_agent_schedule, cancel_agent_run, list_user_searches), and the agent/schedule families are very predictable. The deviations are the three creation tools named noun-first (keyword_search, user_search, expert_search) instead of create_*, which breaks the pattern in the single most-used operation.
30 tools is heavy and crosses the point where surface area itself becomes a selection cost. The count is partly justified by four sub-domains (searches, agent runs, schedules, integrations/usage), but the search family alone spawns ~12 tools that could be consolidated with a resource parameter.
Agent runs and schedules have full lifecycle coverage (create/get/list/update/delete/cancel/rerun; create/get/update/delete/pause/resume/runs), and searches cover create/get/list/paged posts plus polling semantics. The gap is that no search type (keyword, user, expert) can be deleted or updated, and there is no way to re-fetch a stale search beyond retrying.
Maintenance
Related MCP Connectors
Search and read public data from X/Twitter, Reddit, YouTube, Instagram, TikTok, Facebook, Threads, Bluesky, Trustpilot, Amazon, Google Maps, news, forums and the web. 90+ read-only tools plus ready-made research playbooks (lead finding, brand monitoring, review mining). Hosted: connect with an API key; 50 free requests per endpoint each month.
Live public data from TikTok, LinkedIn, YouTube, X, Reddit and 17 more networks, via one key.
Instagram, X, TikTok, YouTube, Facebook and Reddit data. Approved OAuth or API-key access required.
8 social listening tools over one MCP endpoint: Reddit, X, Threads, YouTube, Google Trends, news.
Related MCP Servers
AlicenseAqualityCmaintenanceSocialCrawl provides the "best" quality public web data on the internet. Access 65+ real-time social media, e-commerce and public web data from a single API with a single schema.5569 npm28MIT- AlicenseNot gradedqualityDmaintenanceReal-time X/Twitter social intelligence for AI agents and developers, powered by Grok's live search capabilities.16 npm3MIT

insightsocialofficial
AlicenseNot gradedqualityBmaintenanceEnables AI agents to search, describe and call 239 social data endpoints across Instagram, TikTok, LinkedIn, Facebook, YouTube, X, Reddit, Threads and Pinterest through a single API key, covering profiles, posts, comments, followers, hashtags, ads and transcripts with one unified schema. Full responses are saved server-side and can be re-sliced for free with jq, fields or size outlines, so agents pull only the data they need into context.121 npmMIT
SocialToAI MCP Serverofficial
AlicenseNot gradedqualityCmaintenanceEnables AI tools to read publicly visible social media posts, comments, profiles, and trends from 18 platforms such as TikTok, Instagram, X, Reddit, YouTube, Xiaohongshu, and Weibo, on a prepaid pay-per-call basis where failed calls are free. Ships tools for search, trending, detail, comments, user profiles and posts, plus higher-level workflows like brand buzz, comment analysis, account watching, creator discovery, bulk link data, and viral finding.MIT