For teams running AI agents

FalaAI MCP: transcription, diagnosis and risk audit inside your AI assistant

Same engine as the FalaAI API, a different contract: instead of endpoints, your assistant discovers the tools, picks which one to call and gets the result ready for context. Connect a URL and a fai_ token — no integration code.

5 toolspt-BR, en-US, es-EShealth and dry_run freesame Bearer as the REST APIZero-Storage

From $33.33/month for 90,000 credits

Start freeConnect nowSee the flows

Made byAction Tec Brasil

MCP server endpoint

https://mcp-falaai.action.tec.br/mcpcurl -fsS https://api01-falaai.action.tec.br/v1/health

health needs no key and costs no credits. Analysis tools use the same Bearer fai_ as the REST API.

Connect your MCP client

Each client uses a different field for a remote server. Copy your block and replace fai_... with your key.

Claude Code

File: .mcp.json (raiz do projeto) ou ~/.claude.json

{
  "mcpServers": {
    "falaai": {
      "type": "http",
      "url": "https://mcp-falaai.action.tec.br/mcp",
      "headers": { "Authorization": "Bearer fai_..." }
    }
  }
}
Claude Code MCP

VS Code / Copilot

File: .vscode/mcp.json

{
  "servers": {
    "falaai": {
      "type": "http",
      "url": "https://mcp-falaai.action.tec.br/mcp",
      "headers": { "Authorization": "Bearer fai_..." }
    }
  }
}
VS Code MCP configuration

Cursor

File: .cursor/mcp.json ou ~/.cursor/mcp.json

{
  "mcpServers": {
    "falaai": {
      "url": "https://mcp-falaai.action.tec.br/mcp",
      "headers": { "Authorization": "Bearer fai_..." }
    }
  }
}
Cursor MCP

Gemini CLI

File: ~/.gemini/settings.json

{
  "mcpServers": {
    "falaai": {
      "httpUrl": "https://mcp-falaai.action.tec.br/mcp",
      "headers": { "Authorization": "Bearer fai_..." }
    }
  }
}
Gemini CLI MCP servers

Each client doc is linked under the block. UIs change between versions: what matters is the URL and the Authorization header.

Five tools, one Bearer

The assistant picks the tool; you do not write HTTP. The fields below are the MCP server contract.

health

Free

Checks the connection and returns version, uptime, database and API phase.

Required
-
Optional
-
Returns
status, version, uptime, database, phase, launch_date

transcribe_audio

Turns audio into a transcript with diarization and audio events.

Required
file_base64
Optional
filename, model, language, client_reference_id
Returns
text, dialog (diarizado), audio events

diagnostic

Summary, reason, action, topic and sentiment.

Required
language + (dialog | text)
Optional
duration_seconds (obrigatório quando há dialog), response_language, participants, call_direction, audio_events, client_reference_id
Returns
summary, reason, action, topic, sentiment

risk_audit

Risk and compliance audit with score, detections and an HTML report.

Required
dialog, duration_seconds, language, response_language
Optional
participants, call_direction, audio_events, response_format (v2), client_reference_id
Returns
score de risco, veredito, deteccoes, relatorio HTML

extract_whatsapp

Segments a WhatsApp export into per-speaker conversations; dry_run estimates without charging.

Required
file_base64, timezone, date_format
Optional
filename, start, end, windows, gap_minutes, min_messages, voice_note, expression, dry_run, client_reference_id
Returns
conversas segmentadas, falantes, contagens, usage

Flows with an agent

From the 2-minute test to the file scenario. Every card shows the real path, the effort and the official documentation of the MCP client.

To use the MCP services, connect your client first.

Open quickstart
2 minutesTested in production

AI assistant with MCP

Connect FalaAI's remote MCP server to Claude Code, Cursor, VS Code or Gemini CLI and call the five tools from the chat itself: health (free, no key), transcribe_audio, diagnostic, risk_audit and extract_whatsapp. Audio comes back transcribed with speakers identified; the analysis comes back with a diagnosis, a risk score and traceable HTML — without writing a single line of HTTP. For teams that already work inside an MCP client.

Claude Code MCPVS Code MCP configurationCursor MCPGemini CLI MCP servers
Action implements it
2 minutesScenario

Telegram bot

The bot receives the audio, downloads the file via getFile (Telegram's own 20 MB limit — FalaAI's API accepts up to 1 GB) and returns a speaker-identified transcript via transcribe_audio, a diagnosis (diagnostic) and a risk audit (risk_audit). For auditing field calls, voice-note sales and support conversations sent by your team.

Telegram Bot API — getFile
Action implements it
1 hourDocumented

WhatsApp export

For auditing a WhatsApp export: run extract_whatsapp with dry_run=true to preview the conversations, with no AI and no charge; then split the export into conversations and run diagnostic and risk_audit on each one. Built for batches of support, quality and compliance conversations.

WhatsApp Help Center — export your chat history
Action implements it
1 hourDocumented

WhatsApp voice notes

WhatsApp exports with voice notes: ask for a dry_run=true preview to see conversations, audio minutes and estimated credits, with no AI and no charge; then extract with voice_note=true, which transcribes voice messages inside extract_whatsapp itself, and summarize each conversation from the response. For quality and support teams that receive WhatsApp exports with audio.

WhatsApp Help Center — export your chat history
Action implements it
2 minutesScenario

Audit a recordings folder

Ask in natural language and the agent walks a local folder, transcribes each audio with speakers identified (transcribe_audio) and audits the risk of each call (risk_audit), returning a table with file, score and verdict. Reading the folder is MCP client code, not a FalaAI tool — the client needs access to the directory. For auditing batches of field recordings, call center and quality calls in one run, with each file's JSON kept as evidence.

Action implements it
2 minutesScenario

Complaint triage

Ask for the triage in natural language and the agent walks the batch of conversations the MCP client can read, runs diagnostic once per conversation (dialog, duration_seconds and language) and returns summary, reason, action, label and sentiment. It then filters negative sentiment and delivers the complaint list with reason and suggested action. There is no batch tool: N conversations = N diagnostic calls (plus N transcribe_audio when audio is involved).

Action implements it
2 minutesScenario

Collections and compliance

For collections compliance and quality: the agent transcribes the recording with speakers identified (transcribe_audio) and runs risk_audit with call_direction=outbound and response_format="v2", flagging any approach outside compliance. The response carries verdict, scores, detections, summary and html_report as base64 gzip for traceable evidence. For auditing active collections calls, negotiation scripts and objection handling without writing HTTP.

Action implements it
2 minutesDocumented

Cost before processing

Ask for the cost before processing: the agent calls extract_whatsapp with dry_run=true and returns the whatsapp_preview with credits_total, params, windows, items and warnings, without calling AI and with credits_charged=0. You tune gap_minutes, min_messages and the window, approve the preview and only then run the real extraction, which returns the conversations and usage.credits_consumed. The export goes as file_base64 to the MCP server.

WhatsApp Help Center — export your chat history
Action implements it
1 hourScenario

Summary in the CRM or ticket

The call audio goes through transcribe_audio, diagnostic returns summary, reason, next action, label and sentiment, and the agent logs the note on ticket #123. The CRM is another MCP server (HubSpot, Zendesk or Odoo) connected in the same client: it is not a FalaAI tool and the note is written by the CRM tool, with its own OAuth. It requires an extra MCP connection and someone present to authorize it.

HubSpot — Remote MCP server
Action implements it
2 minutesScenario

Integration health monitor

The agent calls health, which is public, needs no key and no credit, and checks status, version, uptime, database and phase before any charged call. Running it hourly and warning when the API or database goes down is not the MCP's job: the schedule lives in your agent or cron, and the alert in Slack, email or a status page. It is the smoke test before the batch.

Action implements it
2 minutesScenario

HTML report for the client

Compliance needs evidence, not a summary: the agent runs risk_audit on the diarized dialog and returns verdict with level, scores, detections, summary and html_report (base64 gzip). Ask for the decode and the HTML becomes the attachment the client opens in the browser, with the line that triggered each detection. The report only exists with dialog input: plain text returns no html_report.

Action implements it
2 minutesScenario

Multilingual operation

One call, two languages: language describes the conversation, response_language describes the result. The agent transcribes the audio in Spanish (transcribe_audio) and runs risk_audit with language="es-ES" and response_language="pt-BR": the verdict, the detection labels and the HTML come out in Portuguese. Accepted: pt-BR, en-US and es-ES. For auditing a Spanish-speaking operation and delivering the report in Portuguese without manual translation.

Action implements it

Need a custom integration?

Action implements it

FalaAI MCP × closed plugin × pasting JSON into the prompt

Three ways to give your assistant conversation intelligence — and why MCP tools return evidence instead of loose text.

Where it runs

Locked to one product or becomes a manual prompt

Inside your client (Claude Code, VS Code, Cursor, Gemini CLI)

What you write

Hand-built JSON payload for every conversation

No HTTP: the assistant picks the tool and builds the call

Output

Free text that changes on every response

Fixed-schema JSON: turns, detections, risk score and evidence

Proof

No auditable report

Forensic HTML report with quotes and timeline

Cost

Per-seat subscription or opaque credits

Pay per use, with usage.credits_consumed on every response (health is free)

Data

Depends on the plugin vendor

Stateless and zero-storage: no audio, text or result retained

Connect in 2 minutes Talk to Action
What the AI analyzes

What FalaAI Studio identifies in every conversation

Three closed, consistent axes — sentiment, label and action — mean understanding how the customer felt, what they talked about and how the interaction ended. They are FalaAI Studio's base read and do not change the score: they enrich the analysis, not the rating.

FalaAI Studio applies sentiment analysis, contact classification and outcome detection to every contact center interaction. With this closed vocabulary, quality, compliance and operations teams read thousands of conversations in a standardized way: customer sentiment, the contact label (negotiation, cancellation, complaint, support and more) and the final action (resolved, escalated, abandoned, among others). It is transparency about what the engine delivers — for quality monitoring, risk prioritization and operational intelligence.

Sentiment

Dynamic mapping of the customer's polarity and emotional tone across the entire journey — from full satisfaction to critical friction, enabling real-time identification of micro-shifts in mood and hostility.

Label

Automatic identification of the central vector of the customer's intent, classifying the reason for contact through a closed, robust taxonomy built for high-complexity operations.

Action

Intelligent detection of the interaction's closing status, mapping whether the request was solved, retained, transferred or left unresolved by the agent.

Sentiment, label and action can be customized in a tailored plan, specific to your operation.

Proprietary taxonomy

Proprietary taxonomy by risk and quality type

FalaAI's structured forensic taxonomy: a proprietary matrix divided into dozens of macro axes and dozens of behavioural and legal sub-criteria. The engine maps from critical compliance deviations to advanced operational quality patterns, delivering literal evidence and calibrated diagnostics for high-demand operations.

FalaAI applies a proprietary forensic taxonomy for contact-center risk auditing: dozens of macro axes and dozens of sub-criteria with legal and behavioural granularity, with literal evidence per turn. It is the closed ruler that guarantees the same reading across 100% of conversations — no sampling, no subjective judgement — for quality monitoring, risk prioritisation and operational intelligence.

Categories and subcategories can be customized in a tailored plan, specific to your operation.

Credits, not seats

You pay for what you use — minutes of audio or thousands of characters. No per-seat license.

ServiceCost (credits)

Audio transcription

per minute of audio · Minimum 10

60

Diagnosis

per 1,000 characters · Minimum 15

30

Risk Audit

per 1,000 characters · Minimum 90

90

WhatsApp import

per imported conversation

0.15

See plans and pricing

You track usage in the dashboard, upload by upload. Alerts warn you before it runs out.

Plans and Pricing

Monthly credits. If you run out, buy standalone packs that never expire.

What are credits? Learn more →

Free API

Free

Credits/month
2,000one-time credits
Concurrent analyses
1
SLA
99.9%
Start Free

Starter API
90K

$33.33/month

Credits/month
90K
Concurrent analyses
3
SLA
99.9%
Subscribe
Most popular

Creator API
270K

$93.75/month

Credits/month
270K
Concurrent analyses
4
SLA
99.9%
Subscribe

Scale API
2.7M

$833.33/month

Credits/month
2.7M
Concurrent analyses
8
SLA
99.9%
Subscribe

Want a custom plan?

More conversations analyzed at once or specific requirements? We'll tailor it.

Credits that never expire

If the monthly balance runs out, buy standalone credit packages. They never expire and are only consumed after the plan's balance.

No subscription

Try FalaAI Studio for free - Free plan

When you create your account, you get 2,000 free credits - no monthly recharge and no expiry. If you need more, buy as many credit packs as you want, whenever you need - they never expire.

Start Free

Packs by plan

The price per credit drops as the plan grows - so the same pack can have different prices.

Free API

Available standalone credit packages (price per credit varies by plan):

9,000 credits$10.66

Standalone never expires. Consumed after the monthly balance. Available in the dashboard.

Buy

Starter API - 90K

Available standalone credit packages (price per credit varies by plan):

18,000 credits$13.33
45,000 credits$26.32

Standalone never expires. Consumed after the monthly balance. Available in the dashboard.

Buy

Creator API - 270K

Available standalone credit packages (price per credit varies by plan):

27,000 credits$17.65
54,000 credits$28.57
90,000 credits$40.00

Standalone never expires. Consumed after the monthly balance. Available in the dashboard.

Buy

Scale API - 2.7M

Available standalone credit packages (price per credit varies by plan):

270,000 credits$103.45
540,000 credits$181.82

Standalone never expires. Consumed after the monthly balance. Available in the dashboard.

Buy

Credit calculator

Transcription: audio minutes/month. Diagnostic/Audit: enter in thousands of characters (e.g. 1 = 1,000 chars, 5 = 5,000).

Choose a plan

What fits in the plan

60cr/min
min/month
30cr/1000 characters
0 characters/month
90cr/1000 characters
0 characters/month

Reference: 1 min ~ 800 characters · 5 min ~ 4,000 · 10 min ~ 8,000

Use as a guide to estimate thousands above.

Total credits needed

0 credits/month

Fill in the fields above to see the recommendation

Security and compliance

Zero content retention + encryption + LGPD.

Zero-Storage

Audio, text and results are not retained: they are processed and discarded. We keep only usage and audit records — never conversation content.

Encryption

Account data protected in transit (TLS) and at rest, with full per-account isolation. Conversation content is not stored (Zero-Storage).

LGPD

You are the data controller; Action Tec Brasil acts as processor. Data-subject requests are handled within 15 days.

Auditable evidence

Forensic HTML report with a timeline, evidence and violated frameworks — ready for audits and legal disputes.

Open sample Diagnosis reportOpen sample Risk Audit report
See:View LGPD policyView privacy policyView DPA (data processing)View SLA (service level)View platform status

Infrastructure on certified providers

  • Hetzner · ISO/IEC 27001:2022 · BSI C5 Type 2
  • Supabase · SOC 2 Type 2
  • Transcription · SOC 2 · ISO/IEC 27001 · ISO/IEC 42001 · LGPD

Frequently asked questions

An MCP server that exposes five FalaAI API tools (health, transcribe_audio, diagnostic, risk_audit and extract_whatsapp) to AI assistants.
Not for health: it is public and consumes no credits. Analysis tools require the Bearer fai_.

Put your assistant to audit conversations

Start with health, which is free. If the audio sits in the PBX or carrier, Action builds the adapter.

Start freeOpen the dashboard