Loading Ralix…
Point a remote-HTTP-capable MCP client (Claude Code or your own agent) at Ralix's MCP server and call any of the 15 tools below. Each API key is scoped to the exact tool names it may call — a key that only lists get_card can't call run_search.
Endpoint: https://ralix-harness-api.hello-ferlyai.workers.dev/mcp. Create a scoped key in dashboard Settings on any paid plan. Grant only the scopes your assistant needs; start with list_research_runs, list_research_accounts and get_research_account. Keys with older scopes do not gain new permissions automatically. Agents can discover the server from its machine-readable server card or llms.txt.
Create a scoped Ralix key in Settings. Store it in a private local environment variable named RALIX_API_KEY. Keep the actual key out of repositories, chat messages, screenshots and shell history. Add this project configuration to .mcp.json; it contains an environment reference, not the key:
{
"mcpServers": {
"ralix": {
"type": "http",
"url": "https://ralix-harness-api.hello-ferlyai.workers.dev/mcp",
"headers": {
"Authorization": "Bearer ${RALIX_API_KEY}"
}
}
}
}Restart the client with the private environment available, then check /mcp. Other clients must support remote HTTP and a bearer Authorization header; their secret-storage setup can differ. See the official Claude Code MCP reference.
list_research_runs with {"offset":0,"limit":10}.list_research_accounts with a returned runId, offset: 0, limit: 10 and optional name query or decision. Preserve each business's key.get_research_account. Cite its source links and preserve unknowns.research_account only when needed. It returns a durable run ID; poll stream_cards or the dataset instead of starting another run.Example request to your assistant: “Review my local service-business matches against my website service offer. Show the evidence and unknowns. Save a shortlist; do not start another paid search or send messages.”
Authentication failures require a current unrevoked key and paid workspace access. Missing-scope errors require an appropriately scoped key. A review conflict requires re-reading the current revision. Budget stops preserve completed work; do not automatically increase budgets or create fresh IDs to evade a stop. Transient admission failures should be retried with the same requestId. Research availability can remain closed during rollout.
Your Claude or other assistant subscription covers that client's usage, not Ralix's search/model costs. Ralix research uses your workspace allowance. MCP tools do not perform automatic outreach. Read the data dictionary and usage guide.
list_research_runs
Find your workspace datasets and run statuses, in pages of up to 20.
list_research_accounts
Filter all businesses in a run by name or your decision; offset/limit pages of up to 20. Follow nextOffset until null.
get_research_account
Inspect one business by runId and key, including available evidence, provider records, unknowns and your decision.
review_research_account
Save, qualify or dismiss a business with expectedRevision. A stale revision returns conflict; read again before applying a new decision. Your decision never proves automated qualification.
research_account
Enqueue a new paid investigation of a retained business. Supply runId, key and requestId. Retry uncertain admission with the same requestId; use a new one only for another intentional investigation. Existing budgets apply; no outreach is sent.
create_offer
Create an offer — what you sell, your ICP, fit criteria.
run_search
Start a run against an offer, the same pipeline the dashboard's "run" button triggers.
stream_cards
A run's status and the cards it's produced so far.
list_opportunities
Page through visible opportunities, with optional research-run filtering and a continuation cursor.
get_outcome
Read an opportunity stage and its paged, attributable activity history.
update_outcome
Record reviewed, contacted, meeting, won or lost outcomes. Version checks prevent overwrites; request IDs make replay safe.
get_card
Read one already-shipped card in full.
resolve_contact
The contact off an already-shipped card — Ralix resolves contacts inline during a run, there's no separate mid-run resolution step to call.
watch_signal
Save a signal spec for the monitor to re-check on a schedule (the same watch the dashboard's Signals page saves).
schedule_followup
Schedule a reminder against a card — writes the reminder row; sending stays a human (or the dashboard's own send action).
Ralix finds them. Your agents follow through. You close.