Replace your Vapi + n8n + Supabase stack.
One runtime. Every channel.

Agent runtime, workflows, tables with vector search. One config for phone, web, chat, SMS and email.

$curl -H "Authorization: Bearer ak_…" https://api.anyreach.ai/core/agents

REST API · MCP · SOC 2 TYPE II · HIPAA BAA

The same platform behind our BPO and enterprise deployments at anyreach.ai. One runtime, one codebase, now with an API key.

Credit unionsE-commerceHealthcareTelecomHome servicesBPOs

Most agents run on a stack like this.

Three vendors, three dashboards, a webhook mesh you maintain. Every tool call crosses two networks. Every price is per vendor.

vapi

Voiceorchestration$0.05 / min
+ STT + LLM + TTS + telco10 lines, then $10 each

n8n

Workflowsglue$50 / mo + hosting
retries you writecron · http · code nodes

supabase

Tables+ pgvector$25 / mo + sync jobs
embedding pipelineyou keep it in sync

anyreachone API · one config

Agentsphone · web · chat · sms · email

Workflowstriggers · branches · code · wait

Tablesrows · vectors · sync

one billone runtime

It's built for your terminal.

Add the MCP server to Claude Code or Cursor. Agents, tables, workflows, credentials, swarm tests. The dashboard is optional.

claude ~ anyreachbuild 0%

Build an order-status agent for hearthandflame.com. Sync Shopify orders into a table, add a refund workflow with approval over $500, deploy to phone, chat and email, run 20 swarm calls with different personas and show me what failed.

credentials.create shopify · oauth2 · refresher template

tables.create orders · 10,412 rows · embeddings synced

workflows.create refund · 6 steps · approval gate > $500

agents.create order-status v1 · phone + chat + email

ivr_swarm.run 20 personas · goals · against +1 415 ··· 0142

18 pass · 2 fail · refund over $500 skipped approval

swarm 07 · "refund the oven" · $612 · approval step never reached

workflows.update refund v2 · approval before shopify.refund

ivr_swarm.run 20 / 20 goals met

agents.publish v2 · +1 415 555 0142 · widget · support@hearthandflame.com

›

Settings → Connectors → Add custom connector
URL: https://api.anyreach.ai/mcp
→ sign in with your Anyreach account. No client ID, no secret.

// .cursor/mcp.json  (Cursor needs the published client ID)
{ "mcpServers": {
    "anyreach": { "url": "https://api.anyreach.ai/mcp",
                  "clientId": "see docs.anyreach.ai/mcp" }
} }

Any client that supports remote servers + Client ID Metadata
Documents connects with the URL alone:
https://api.anyreach.ai/mcp

Your account, your permissions, every org you belong to. Two tools, search_tools and call_tool, reach the whole API.

Every operation and three worked sessions →

We built the backend in.

Workflows read and write the same tables the agent queries mid-call. No glue service. No second vendor. The dashboard, the API and the MCP server see one state.

For humans

orders· 10,412 rowsstatusskuemail

order	item	status	match
4471-0928	Karu 16 · cobalt	shipped · Oct 14	0.91
4471-0113	Koda 12 · slate	delivered · Jun 02	0.74
4470-9986	Cover · Karu 16	delivered · Jun 02	0.62

workflow · refundv2

1	trigger · call.ended	done	evt
2	extract · order_id, amount	done	llm
3	approval · amount > 500	waiting · 14m	human
4	shopify.refund	queued	http

For agents

mcp · anyreach

> tables.search("orders", "blue oven I bought in June",
    filter={ email: caller.email }, limit=3)
← 3 rows · 61ms · cosine
  4471-0928  Karu 16 · cobalt   shipped    0.91
  4471-0113  Koda 12 · slate    delivered  0.74
  4470-9986  Cover · Karu 16    delivered  0.62

> workflows.run("refund", { order: "4471-0928", amount: 612 })
← run_01k7 · step 3/6 · waiting approval

Workflows

Triggers, conditions, loops, parallel branches, sandboxed Python, HTTP, wait steps, and app steps for 2,000+ SaaS tools. Callable as tools mid-conversation, or pre- and post-call.

workflows.run · workflows.schedule

Tables + vector search

Rows, not documents. Filter by column, rank by meaning. Sync from Sheets or Shopify. Pair a table with a recrawl workflow and you have a knowledge base that watches your site and proposes instruction edits.

tables.search · tables.upsert · tables.claim

Credentials that refresh

Secrets stored encrypted, referenced from any step. Expiring tokens refreshed by a workflow you write once; ready-made templates for OAuth and the usual providers.

$credential(shopify)

Wherever your customers are, the agent is there too.

Define the agent once. The same version, tools and tables answer the phone, the widget, SMS and the inbox. Every agent is a SIP endpoint, so it drops into Genesys, Five9, NICE CXone, Avaya or any trunk you bring. Every turn becomes an event you can subscribe to.

Channels

Phonesip endpoint · byoc trunk

Webwidget · voice + chat

Chatembed · api

SMS10dlc · otp

Emailthreads

One config

order-status.yaml

agent: order-status
version: 13
model: openai/gpt-4.1
identity:
  verifier: workflow:verify_caller
tools:
  - tables.search: orders
  - workflows.run: refund   # needs verified identity
channels: [phone, web, chat, sms, email]
overrides:
  email.model: openai/gpt-4.1-mini
  phone.tts: elevenlabs/v3

It tells you first

anyreachapp9:41 AM

order-status v13 · last hour, all channels.

212 conversations, 91% contained
Phone p50 412 ms, 2 warm transfers
1 refund waiting approval, $612
Email: 3 threads, same agent, 0 escalations

See it on every channel ↗

One agent. A full day.

Same version, same tables, same tools. Different channel every hour. Nobody rebuilt a prompt.

09:12 AM · Phone

Find the order. No number needed.

order-status · +1 415 555 0142live

callerWhere's my order? The blue oven I bought in June.

tooltables.search("orders", "blue oven June", filter={email}) → 2 rows · 61ms

agentFound it. Order 44-71, 09-28, the Karu 16 in cobalt. Shipped October 14, out for delivery today.

callerEmail me the tracking link?

toolworkflows.run("send_tracking") → queued

agentSent. Anything else?

48s · containedp50 412ms

03:05 PM · Web chat

Verify by SMS. Don't leave the chat.

Hearth & Flameweb chat · hearthandflame.com

Can you change the delivery address on my order?3:05 PM

Sure. To change the address I'll text a code to the number on the order.3:05 PM

Code sent to ···0199.3:05 PM

44713:06 PM

Verified. What's the new address?3:06 PM

verifier workflow passed · identity: verified · tools.update_address unlocked · chat + sms in one conversation

See it on your site. Five minutes.

We crawl the site, build a phone, chat and email agent from it, put it on a number and email you the link. No signup, no form.

Email a URL. Get a demo back.~5 min

to · demo@mail.anyreach.ai

body · the prospect's website, nothing else

cc · anyone who should get the link

reply 1 · received · reply 2 · live demo on a phone number, a widget and an inbox

Open in mail

Any model. Swap it per channel.

Pick the LLM, the recognizer and the voice per agent, then override any of them per channel. A cheaper model on email, a faster one on the phone. No redeploy.

LLM

OpenAI GPT-4o · 4.1 · 5 · 5.x · 6
Azure OpenAI same catalogue
Google Gemini 3 · 3.5 · 4
xAI Grok 4.x
DeepSeek V4 · Kimi K2.6
Gemma 4 · GPT-OSS 120B

STT

Deepgram Nova-3 · Flux · Medical
AssemblyAI Universal
Cartesia Ink
Gladia code-switching
Speechmatics · Google

TTS

Cartesia Sonic 3.x
ElevenLabs turbo · cloning
Deepgram Aura-2
Rime · Inworld · Fish
Google Gemini TTS

Replace your Vapi + n8n + Supabase stack. One runtime. Every channel.