Bridgecore — plain words to n8n workflow (automation planner for people and AI agents)

Say what should happen, in plain words. Bridgecore picks the right tool for each step from 3,597 app capabilities (n8n nodes incl. AI steps, OpenAPI services, MCP servers) and gives you back an importable n8n workflow JSON, the matching ready-made recipes, and the honest list of blanks only you can fill. Or, in do mode, it finds the one capability your sentence needs and makes the call for you.

when a typeform is submitted add a row to google sheets and send a slack message to #leads

→ Typeform Trigger → Google Sheets (append row) → Slack (#leads filled in) · blanks: the sheet · accounts: Typeform, Google, Slack.

No login to any app is needed to plan. Nothing you say is stored beyond the run's own dataset.

What you can ask for

You say	You get
when a new email arrives in gmail summarize it with openai and send the summary to slack	Gmail Trigger → OpenAI (prompt written) → Slack, 3 accounts to connect
when a shopify order is paid text the customer a thank you	Shopify Trigger → Twilio SMS, phone taken from the order
when a github issue is opened classify it with an ai agent and comment	GitHub Trigger → AI Agent (tools attached) → GitHub comment
when a hubspot contact is created send me a telegram message	HubSpot Trigger → Telegram
text +15551234567 saying the report is ready (output: do)	the exact Twilio request, dry-run first; dry_run: false sends it

Works for lead capture, order notifications, alerts and monitoring, social posting, AI summaries and classification, agent workflows — any app n8n, an OpenAPI spec or an MCP server can reach (Google Sheets, Slack, Gmail, Notion, Airtable, HubSpot, Shopify, Stripe, Telegram, Discord, Twilio, OpenAI, Gemini, Claude and hundreds more).

Input

Field	Meaning
needs	One automation per line. Start with when … for the trigger.
output	all (default: recipes + plan + workflow) · recipe · plan · workflow · do
inputs	do mode: values for the capability's inputs by name, e.g. {"to": "+1555…", "body": "hi"}
auth	do mode: headers the call needs, e.g. {"Authorization": "Bearer …"} — used inside this run only, masked, never stored
dry_run	do mode: true (default) shows the exact request and sends nothing; false makes the call

Example input:

{ "needs": ["when a typeform is submitted add a row to google sheets and send a slack message to #leads"], "output": "all" }

Output (one dataset item per need)

• plan.steps[] — text, kind (trigger/action), tool (id, app, summary, inputs, auth) or gap when the named app cannot do that step. A gap is reported, never papered over with another app.
• workflow — importable n8n JSON (nodes wired in order, said values filled). Import it in n8n, connect the accounts, test, turn on.
• blanks[] — what only you can fill (node, need). accounts[] — credentials to connect.
• recipes[] — tested ready-made recipes that match, each with its own workflow JSON; public_templates[] — matching public n8n templates by title and official link.
• do mode returns steps[] instead: per step the matched capability, request (method, url, headers masked, body), missing[] inputs, and when dry_run: false the service's status, data or error; plus ran, charged and fee_usd so you can see exactly what was billed.

For AI agents (Apify MCP, OpenAI/Claude tool use, LangChain, CrewAI…)

This Actor is exposed by the Apify MCP server (https://mcp.apify.com) like every Store Actor. An agent can call it without reading this page:

call-actor  bridgecore/bridgecore-automation-planner
input       {"needs": ["<the user's sentence>"], "output": "plan"}

Use output: "plan" when you only need the right tool per step, "workflow" when the user wants something to import into n8n, "do" when the user wants the thing done now (dry run first, then dry_run: false with the user's own auth). The result is plain JSON; gap entries tell you what to ask the user.

The same engine is reachable without Apify as the MCP server io.github.chinlee2200-dotcom/bridgecore (official MCP registry · Smithery · Glama) with tools do, find_recipe, plan_automation, make_n8n_workflow, and as the free site https://bridgecore.chinlee2200.workers.dev .

Pricing (pay per event — nothing is charged for a need that errors)

Event	When	Price
automation (default)	one need answered with recipes + plan + workflow	$0.05
recipe / plan / workflow	one need answered with just that part	$0.02
run	one successful real call in do mode	$0.05 — dry runs, errors, missing inputs and 4xx/5xx replies are never charged

Compute is included in the event price.

How it decides (so you can trust or doubt it)

• The engine indexes real node descriptions from n8n (565 nodes, 121 AI/agent nodes), OpenAPI specs and MCP tools/list — one capability per resource/operation. Matching is measured: single-step top-1 93 %, top-3 100 %; multi-step "every step right" 61/61 on the held-out set (2026-10-14).
• Values you say out loud (emails, phone numbers, #channels, trigger events) are filled; everything else is left as a named blank rather than guessed.
• Workflows are shape-checked against n8n's own validator, but you test them in n8n before turning them on.
• Public templates are listed by title and official link only; open the link on n8n.io to use them.

FAQ

Is this an n8n template scraper? No. It designs a workflow for your sentence and only points to public templates by title and link. Does it store my API keys? No. auth is used inside the run that received it and masked in every output. Does it need an n8n account? Planning: no. Importing the workflow: yes, any n8n (Cloud or self-hosted). Which apps? Anything n8n has a node for, plus any OpenAPI service or MCP server in the index. If an app cannot do a step, the plan says gap.

Source: github.com/chinlee2200-dotcom/Laboratory- (core/, apify/bridgecore-actor). Issues and requests welcome there.