FuseAI MCP

The Agents / Signals feature: create, list, activate, and pause signal monitors — recurring lead/intent watchers (e.g. companies hiring SDRs, funding announcements, people engaging with a LinkedIn post, job postings). Hand off the user's full request here for anything about signal monitors or 'agents'.

Read-only performance dashboards & metrics: campaign metrics and trends (email / LinkedIn / dialer, with up-or-down deltas), dialer call analytics, prospecting/enrichment volumes, and team performance. Use for 'how is X performing / trending / up or down' questions. NOT the raw call log (that is power_dialer), and NOT for building or editing campaigns (that is campaigns).

Ask the Fuse Brain — living relationship files for ENGAGED contacts/companies, built from the team's real emails, LinkedIn messages, calls and meetings. Best tool for questions about existing relationships: history, needs, objections, commitments, deal state ('what does Dana care about?', 'which contacts mentioned budget?', 'where did the Acme deal stall?', 'who is waiting on us?'). Answers cite the contacts/companies used. Not for cold prospects: no file exists until a contact engages.

Create and manage AUTOMATIONS — scheduled or event-triggered multi-step rules (website/news monitors with Slack digests, follow-up drafts, auto-enrolment into campaigns, CRM saves, AI steps, and rules that listen to or act in the user's connected apps: Google Calendar, Google Sheets, Notion, Typeform, Zendesk, Airtable, Asana, Linear, ...). Route here for anything recurring — 'daily', 'every time', 'whenever', 'monitor' — or any if-this-then-that request ('automate X when Y', 'when a Typeform response arrives...', 'before each calendar meeting...'), for connecting an app an automation needs, for 'I connected <app>, continue', and for inspecting, editing, pausing, resuming, deleting, or asking about the runs of an existing automation / rule. Pass the user's full request (verbatim) as `request`.

Outreach campaigns and their email/LinkedIn/dialer SEQUENCES, end to end: list/create/update/duplicate/delete campaigns; build, generate, edit, reorder, test and preview sequence steps; ATTACH LISTS to a campaign (this is how you set up reaching a list of people, including system lists like All Website Visitors — attaching to an ALREADY-ACTIVE campaign enrolls its contacts in the background, while a draft campaign enrolls at launch); manage leads (pause/resume, activity); templates; and per-campaign performance metrics + trends. NEW campaigns built here stay DRAFTS until the user asks to launch; an existing active campaign keeps running. (For cross-feature dashboards — dialer/prospecting/team analytics — use analytics instead.)

Starts an asynchronous Fuse Deep Research run: over several minutes it researches the request across the web in the background and produces a markdown report, plus — when the request is to find people or companies — a new Fuse prospect list. Runs bill credits. ONLY call this when the user EXPLICITLY asks for "deep research", or confirms they want Deep Research after you offered it. For a general research/find-X request that does not say deep, do NOT call this — use the quick `research` tool, or ask the user whether they want quick research (immediate answer) or Deep Research (background, 5-20 minutes, report + prospect list) and wait for their choice. Returns immediately with a runId; it can NEVER wait for or preview results in the same call. After starting, call deep_research_status with the runId to check progress (suggest the user checks back in a few minutes). Pass the user's full request as `query`, an optional short `listName` when a list is expected, and `effort` if specified.

Checks a Fuse Deep Research run started with deep_research_start. Returns the run status (queued/planning/running/generating_csv/uploading/completed/failed/cancelled) and, once completed, the full markdown research report plus the id of any produced prospect list. Use it whenever the user asks whether their research is done or wants the results.

Use for anything in the user's inbox / messaging: reading, sending, flagging (read/archive/spam/delete), and tagging emails; AI-drafting email or LinkedIn replies; approving/declining/skipping pending auto-replies; and LinkedIn messaging (chats, messages, connections, invitations). Pass a clear, self-contained natural-language request.

CRM & integrations: connection status, CRM Fields/sync settings, owner mapping, pulling/pushing contacts & companies to the connected CRM (HubSpot/Salesforce/Zoho/Pipedrive/Attio), custom contact properties, and Slack notifications. Pass the full task as `request`.

Power Dialer — call history (READ-ONLY). Use for questions about the user's recent calls / call history. Cannot place or manage live calls.

Manage the user's prospect lists: list lists (including the system/auto lists — All Website Visitors, All Saved Contacts, All <CRM> Leads, All Enriched Contacts), view/search a list's rows, create/append/clean lists, smart columns, and resolve a list's real id. This is where you FIND the listId of a cohort (e.g. the people who visited the website) so a campaign can target it. NOT for finding brand-new prospects on the internet or the prospect database (that is prospect_search).

Find people and companies, generate prospect/company search filters from a natural-language description, enrich contacts, and save prospects to lists. Use for anything in the Prospect Search area.

EXTERNAL / internet research and web tasks: search & scrape the web (Firecrawl), Google Maps & Google Jobs lookups, building a prospect list from the web (EXA websets), internal knowledge-hub semantic search, and FuseAI how-to / navigation help. NOT for searching FuseAI's own prospect database — that's prospect_search. Examples: 'find recent funding news about Acme Corp', 'build me a list of AI infra startups in SF from the internet'.

Account & workspace settings: company profile (name/website), enrichment preferences, team members & invitations, Knowledge Hub profiles, writing styles, ICPs (Ideal Customer Profiles), campaign visibility, a thin lifecycle view of existing Rules Engine automations (list/pause/resume/delete only — it does NOT create or edit automations), Deep Research provider keys (BYOK), and MANAGING the connected sending accounts — email accounts (settings/sharing/disconnect/SMTP connect/warmup start-pause), LinkedIn accounts, dialer phone numbers, and purchased sending domains & mailboxes. Also the door for API keys and billing/subscription questions (it can READ the current credit balance; everything else billing it explains is app-only). NOT for CRM connection/sync/owner-mapping. Pass the full settings task as `request`.

The Smart Actions feed — the home list of suggested next actions per contact, aggregated from website visits, signal-monitor matches, campaign replies, and ICP matches. Use it to answer 'what should I do next', to act on website-visitor or signal items one contact at a time (inspect -> enrich -> send), and to triage/execute the feed. Examples: 'what should I do next?', 'send the top suggested action', 'work through my website-visitor actions'.

Use for Website Intent READ-ONLY reports + tracking config: who (companies/people/industries) is visiting the website, visitor analytics/trends, and the tracking setup (allowed domains, tag script, de-anonymization credits). NOT for acting on visitors — to turn website visitors into a list, reach them, or add them to a campaign, use prospect_lists (the 'All Website Visitors' list) + campaigns. Examples: 'which companies visited my site this month?', 'show my visitor trend', 'is my tracking tag live on acme.io?'.