Inbound leads, researched and routed before you finish your coffee

Why now

• n8n launched Agents (Sep 25, 2026, in preview on n8n Cloud). You describe what the agent should do, give it a model and tools (including MCP servers and your own workflows), and mark sensitive tools for human approval. A workflow can call it with the new Message an Agent node, and each agent turn counts as one execution. (n8n blog)
• Attio's MCP server is on every plan, Free included, at https://mcp.attio.com/mcp with OAuth. (Attio MCP docs)
• Tally webhooks are free on all plans, with optional SHA256 request signing. (Tally webhooks)

Together, the fixed parts (receive, verify, notify) stay as a predictable workflow, and only the open-ended part (research and judgment) goes to an agent.

Who it's for

Small teams whose inbound arrives through a "work with us" or "book a demo" form and goes to a shared inbox to die. It assumes Attio as the CRM, but the shape works with any CRM that has an MCP server.

What you get

Within a few minutes of a form submission:

• The company and person found or created in Attio, with a research note: what they do, size signals, why they might be writing, and a fit score against your criteria, each with its source link.
• A Slack message in #inbound: name, company, a 3-line brief, the fit score, and a link to the Attio record.
• Approval before writes: the agent pauses for Approve or Reject before it creates or changes anything in Attio.

You'll need

Setup

1. Build the form (Tally, 10 min)

Fields: Name, Work email, Company website, What are you looking for? (long text), Budget range (dropdown), and a required consent checkbox: I agree you can research my company using public sources to route this request. Publish it.

2. Write your fit criteria (any doc, 10 min)

The agent needs them in writing. For example:

Great fit (score 4–5): B2B software, 20–500 employees, needs <your service>, budget ≥ $10k.
Maybe (2–3): right industry but small or vague.
Not a fit (0–1): students, agencies reselling, job seekers, vendors pitching us.

3. Create the agent (n8n, 30 min)

In n8n Cloud, open Agents → Create Agent and name it Inbound researcher.

• Model: any capable model you have credentials for, or Gateway credits.
• Tools:
• Attio as an MCP server (https://mcp.attio.com/mcp, OAuth). Exclude every tool except search, create and update for companies, people and notes. Mark create and update as requires approval.
• A web search or HTTP Request tool for reading the company's website.
• Slack: a Slack node set up for one action only: post to #inbound.
• Instructions:

You research inbound leads and route them. Input: a form submission (JSON).
1. Read the company website (homepage, pricing, about). Use public sources only.
   Never look up personal social profiles.
2. Score fit 0–5 using these criteria: <paste step 2>. Explain the score in one line.
3. In Attio: search for the company by domain and the person by email. If they exist,
   add a note. If not, create them, then add a note. Note format:
   "Inbound <date>: <what they asked for>. Research: <3 bullets with source URLs>.
   Fit: <score> – <reason>."
4. Post to #inbound: "*<Name>* at *<Company>* · fit <score>/5
   <3-line brief> · <Attio record link>". Nothing else, nowhere else.
If the website can't be read, say so and score "unknown". Never guess revenue.

Test it in the agent's chat with a fake submission before you connect anything.

4. Build the workflow around it (n8n, 30 min)

Create a workflow:

1. Webhook trigger (POST). Copy the production URL.
2. Code node: verify the Tally-Signature header, an HMAC-SHA256 of the body with your signing secret (base64), and stop if it doesn't match. Then flatten Tally's data.fields array to {label: value}.
3. Message an Agent node: agent Inbound researcher, message New inbound lead: {{ JSON.stringify($json) }}.

Activate the workflow.

5. Connect Tally (Tally, 5 min)

In the form's Integrations tab, choose Webhooks → Connect, paste the n8n production URL and add a signing secret (the same one used in step 4). Submit a test entry and check the Tally event log, the n8n execution and the agent session.

6. Supervise the first 20 (a week of light checking)

Approve or reject each Attio write in n8n as it comes in. Tighten the instructions where the scores feel off. When the notes are consistently good, you can remove approval from note creation and keep it on record creation.

Time to set up

About 2 hours (estimate, not timed), mostly the agent's tool setup and testing with fake leads.

Cost

• Minimum: n8n Cloud Starter €20/mo (annual; €24 monthly) + model usage, with Tally Free, Attio Free and your existing Slack. Each lead uses a handful of executions: the workflow run plus one per agent turn.
• Model usage is separate. A research-and-note run is a few cents with mainstream models (an estimate; check your provider's pricing).
• Growing team: Attio Plus at $35/seat/mo (annual) for more seats, reports and credits.

Prices are list prices as of Oct 14, 2026.

Gotchas

• n8n Agents is a preview. n8n says agents work but are improving release by release. Test before publishing, and keep approvals on anything that writes.
• Tally waits 10 seconds. The webhook must answer within 10 seconds or Tally retries (at 5 min, 30 min, 1 hr, 6 hrs, 1 day). Respond immediately in the Webhook node ("Respond: immediately") and let the agent run after.
• Signature check needs the raw body. If the HMAC never matches, you're probably hashing re-serialized JSON. Hash the payload exactly as received.
• MCP gives broad access. Connecting Attio's MCP server exposes all its tools unless you exclude them. Exclude deletes and anything list-wide.
• Attio Free credits. AI features in Attio (Ask Attio, workflows) spend credits, but this chain does the AI work in n8n, so it doesn't depend on them.
• Privacy. Research the company, not the person. Keep the consent checkbox, and don't enrich personal data you haven't been given.

Remix

• Calendar-first: route 4–5 scores to a Slack message with your booking link, ready for you to send.
• Swap the CRM: HubSpot or Notion with their MCP servers. The instructions barely change.
• Weekly digest: schedule the same agent on Friday to summarize the week's inbound by fit and source.

Freshness log

• Oct 14, 2026: First published. Desk-checked against n8n's Agents launch post and pricing page, Attio's MCP docs and pricing page, and Tally's webhooks guide and pricing page.

Last verified: Oct 2026 (desk-checked, not run end-to-end). We checked every step, setting and price against the vendors' current docs. We have not yet run this chain live. If a step doesn't match what you see, reply and tell us. It'll be fixed in the next refresh.