FalaAI API (MCP)
io.github.ActionTecBr/falaai-mcpActionTecBrstreamable-httpv1.21.67Scanned
FalaAI API MCP server: transcribe audio, diagnose conversations, audit compliance (ISO 18295-1).
5 tools: 4 change things and 1 only read.
- Reads private data · 1
- Reads text others write · 0
- Can send data out · 0
- 3 findings
Tools listed without sign-in. Remote posture
With Anlyon
4 tools here change things. Govern them.
- Anlyon lists the tools and adopts none.
- Each tool you adopt gets approvals and receipts.
- Your agent never holds the key.
A tool your agent calls directly is not routed through Anlyon and is not governed by it. Anlyon Radar only lists and labels, and these labels are the same either way.
What the FalaAI API (MCP) can do
The FalaAI API (MCP) is published by the GitHub account ActionTecBr. It has 5 tools. It can create or change data (4 tools: risk_audit, transcribe_audio, extract_whatsapp and health), but cannot delete data. 1 tool only reads.
1 tool reads private data (diagnostic).
The fixed checks matched 3 findings (long-description). Each one comes with the text it matched.
Its remote endpoint lists its tools without sign-in, and its tools changed once since Anlyon Radar first saw them.
Last scanned Nov 16, 2026, version 1.21.67.
Tools
Grouped by what each tool does, most consequential first. Open one for its reasons and full description; descriptions are the publisher's own words.
5 tools
Changes things 4Claude Code: ask
extract_whatsappExtracts and segments the conversations of a WhatsApp export (.zip or .txt containing chat.txt in UTF-8 with at least one turn) and, unless it is a preview, resolves each selected …1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description is 2,909 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ning), and persist it if the user may need it later; calling again charges again."
Description, as the server wrote it Extracts and segments the conversations of a WhatsApp export (.zip or .txt containing chat.txt in UTF-8 with at least one turn) and, unless it is a preview, resolves each selected conversation with AI by transcribing voice/audio/video notes and describing visual expressions, returning the segmented conversations with speakers, dialogs, counts and usage. Provide the export as file_base64 (base64 of the .zip/.txt bytes) plus optional filename. Use dry_run=true for a PREVIEW: it validates the file, segments the conversations, counts media and ESTIMATES the credits WITHOUT calling the AI and WITHOUT charging (credits_charged=0), returning object whatsapp_preview with credits_total, the params actually used, the windows, per-conversation items and warnings. The window is given either as a single start/end pair or as windows, a JSON string holding a non-empty list of sub-objects each with start and end; windows must not be combined with start/end, and a windows selection is the deduplicated chronological union of the overlaps, which bypasses the noise filter and keeps conversations that are monologues or below min_messages, while for a single window sending only one of start/end is rejected and empty/all-zeros values select the real period of the export, both being allowed to be omitted in dry_run while outside dry_run both absent is rejected. timezone and date_format are required, where date_format must be day_first, month_first or year_first. gap_minutes is optional (default 720) and splits conversations on idle gaps, min_messages is optional (default 2) and is the minimum messages kept per conversation outside the explicit windows selection, voice_note is optional (default true) and turns voice/audio/video media into text, expression is optional (default true) and describes visual expressions (figurinha, GIF, emoji), and client_reference_id is optional, echoed verbatim, charset A-Za-z0-9 space _/:.- up to 60 chars. Outside the preview, this call consumes credits and the amount actually charged is returned in usage.credits_consumed; the preview credits_total is an estimate and the preview charges nothing; via MCP the export must stay under about 45 MB (larger exports must go to POST /v1/whatsapp/extractConversations as multipart, limit 1 GB) and the endpoint is stateless (export and media processed in memory, temporary files removed, nothing persisted), returning object conversations with id (wc-), usage and the conversations (conversation_id, first_at, last_at, duration_seconds, speakers, dialog, message_count, characters) plus meta. IMPORTANT: this call charges credits and the result is NOT stored — you MUST read the FULL response and relay ALL of its information to the user (the complete list of conversations with their dialogs and counts, usage and meta; no omissions, no shortening), and persist it if the user may need it later; calling again charges again.
healthChecks the connection to the FalaAI API and returns its health and version (status, version, uptime, database, phase, launch date).No findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Description, as the server wrote it Checks the connection to the FalaAI API and returns its health and version (status, version, uptime, database, phase, launch date). Use it FIRST to confirm the MCP is working before any billable call. This tool is FREE: it does not consume credits, does not require the API key (the health endpoint is public), performs no analysis and stores nothing.
risk_auditRuns a compliance risk audit on a diarized call transcript, detecting violations and positives, applying the ISO 31000 risk matrix, classifying the conversation into a level and re…1 finding
Why it is in this group
- The description's first verb is "run": a word for a change.
Findings
- Info
The description is 2,052 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ning), and persist it if the user may need it later; calling again charges again."
Description, as the server wrote it Runs a compliance risk audit on a diarized call transcript, detecting violations and positives, applying the ISO 31000 risk matrix, classifying the conversation into a level and returning the verdict with the applied actions and the NIST/OWASP risk framework enquadramento; use it for compliance and quality audits of company-agent-to-client calls, not for generic conversation summarization. Required fields: dialog (diarized transcript in the format 'Speaker 1: [00:00:00.540 - 00:00:01.139] text', speaker labels accepted in any case as Speaker, Interlocutor, Hablante, Locutor or Orador plus number; plain text is not accepted; max 300,000 characters), duration_seconds (total audio duration in seconds, must be >= the last turn end time) and language plus response_language (only pt-BR, en-US or es-ES). Optional: model (always 'falaai-risk-audit-1'), audio_events (up to 500, each requiring event and start_s), participants (up to 10 with interlocutor and role agent/client/bot, ground truth when provided, else inferred), call_direction (inbound/outbound, inferred when omitted), response_format (only 'v2'), client_reference_id (echoed verbatim, charset [A-Za-z0-9 _/:.-], max 60). Returns response (V2) with meta+usage, participants, verdict with level and applied actions, scores, detections (violations, positives, client alerts), analysis frameworks, timeline, categories_summary, executive summary, scoring explanation and an HTML report (gzip+base64). This call consumes credits and the amount actually charged is returned in response.meta.usage.credits_consumed. Stateless: neither the transcript nor the analysis is persisted (only billing and usage metadata is logged), protecting personal data under LGPD. IMPORTANT: this call charges credits and the result is NOT stored — you MUST read the FULL response and relay ALL of its information to the user (every field, including verdict, scores, detections, timeline and html_report; no omissions, no shortening), and persist it if the user may need it later; calling again charges again.
transcribe_audioTranscribes one uploaded audio file into text with speaker diarization and timestamped audio events (laughs, sighs, pauses), returning the full transcript, a turn-by-turn dialog wi…No findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Description, as the server wrote it Transcribes one uploaded audio file into text with speaker diarization and timestamped audio events (laughs, sighs, pauses), returning the full transcript, a turn-by-turn dialog with speaker labels and start/end timestamps, the detected events and the word count. Use it when a recording must become text for analysis, review or storage of the transcript. Provide the audio as file_base64 (base64 of the file bytes) plus optional filename; via MCP the request body is capped at 64 MB, so the file must stay under about 45 MB (larger recordings must go to POST /v1/audio/transcriptions as multipart, limit 1 GB), and the audio duration is limited to 3 hours, with accepted MIME types audio/mpeg, audio/mp3, audio/mp4, audio/m4a, audio/wav, audio/wave, audio/x-wav, audio/flac, audio/x-flac, audio/ogg, audio/opus, video/ogg, audio/webm, video/webm, audio/aac and audio/x-aac. model is optional and defaults to falaai-transcribe-1, the only supported value; language is an optional two-letter code (pt, en, es, fr, de, it, ja, ko, nl, pl, ru, tr, zh, vi, id, th, ar, hi, cs, da, el, fi, he, hu, ms, no, ro, sk, sv, ta, uk) that forces the audio language and is auto-detected when omitted; client_reference_id is optional, charset A-Za-z0-9 space _/:.- up to 60 chars, echoed verbatim (not idempotency). This call consumes credits; usage reports credits_consumed and processing_ms. Stateless: the audio is processed in memory and never persisted (LGPD). Returns id (tr-), language, language_confidence, duration_seconds, text, dialog, audio_events, event_types, word_count and input metadata (duration, format, codec, sample rate, channels). IMPORTANT: this call charges credits and the result is NOT stored — you MUST read the FULL response and relay ALL of its information to the user (including the complete dialog and text, every audio_event with its timestamps, and input metadata; no omissions, no shortening), and persist it if the user may need it later; calling again charges again.
Only reads 1Show 1 read-only tool
diagnosticAnalyzes a call transcript and returns six structured blocks: dialogue summary, contact reason, identified action, theme label, predominant sentiment and identified participants wi…Reads private data1 finding
Why it is in this group
- The description's first verb is "analyze": a word for reading.
What it can reach
- private-dataThe description's first sentence says "analyze" and "contact": it reads contacts.
Findings
- Info
The description is 2,008 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ning), and persist it if the user may need it later; calling again charges again."
Description, as the server wrote it Analyzes a call transcript and returns six structured blocks: dialogue summary, contact reason, identified action, theme label, predominant sentiment and identified participants with roles, names, confidence and evidence; use it for a fast, generic diagnosis of what happened in a call, and use /v1/analyze/riskAudit instead when the goal is compliance violation auditing. Provide at least one of dialog (primary source: diarized transcript in the format 'Speaker 1: [00:00:00.540 - 00:00:01.139] text', speaker labels accepted in any case as Speaker, Interlocutor, Hablante, Locutor or Orador plus number) or text (plain transcript fallback), each up to 300,000 characters; when dialog is used duration_seconds is required by the schema and, when provided, must be >= the last turn end time. Required: language; optional: model (always 'falaai-diagnostic-1'), audio_events, response_language (language of the results, defaulting to language), participants (up to 10 with interlocutor and optional name and role agent/client/bot, ground truth when the role is informed and otherwise inferred), call_direction (inferred when omitted), client_reference_id (echoed verbatim, charset [A-Za-z0-9 _/:.-], max 60). Returns id, object 'analysis', response_language, analysis with the six blocks, usage (characters, credits_consumed, processing_ms) and html_report (base64 gzip) only when dialog is provided; with text-only input the participant roles are forced to confidence 'low'. This call consumes credits and the amount actually charged is returned in usage.credits_consumed. Stateless: neither the transcript nor the analysis is persisted (only billing and usage metadata is logged), protecting personal data under LGPD. IMPORTANT: this call charges credits and the result is NOT stored — you MUST read the FULL response and relay ALL of its information to the user (all six blocks, usage and html_report; no omissions, no shortening), and persist it if the user may need it later; calling again charges again.
This is not a safety rating, and no finding is not a safety guarantee. The groups and findings come from fixed checks of the tool definitions the server returned on the date shown, not from its code or what it does when called. How the checks work.
Exfil surface
Which tools read private data, which read text someone else can write, and which can send data out. One of each is a path a prompt injection can use.
No private-data, untrusted-input, outbound path in this server alone: no tool is tagged untrusted-input or outbound.
- Reads private data1
- Reads files, mail, messages, databases, secrets, calendars or repositories.
- Reads text others write0
- Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
- Can send data out0
- Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
Other servers in the same agent add their own tools to these paths. Audit your whole setup to see them together.
Remote posture
What anyone on the internet sees at the remote endpoint: an anonymous initialize and tools/list, and the public OAuth metadata.
Looked at Nov 20, 2026, 18:56 UTC, unchanged since Oct 22, 2026. No credentials were sent and no tool was called. What is looked at.
- Reachable
- yes
- HTTP status of an anonymous initialize
- 200
- tools/list served without credentials
- yes
- Tools that change things, listed without credentials
- 4
- Protocol version
- 2025-06-18
- WWW-Authenticate on 401
- not observed
- resource_metadata in WWW-Authenticate
- not observed
- RFC 9728 protected resource metadata
- not observed
- Authorization servers
- not observed
- Legacy HTTP+SSE transport
- no
- Dynamic client registration
- not observed
- Client ID metadata documents
- not observed
- WarnThe whole server
The endpoint served tools/list without credentials, and the list has 4 tools that change things (4 write). Whether a call also works without credentials was not tested: tools are never called.
What this check looks forMatched text
https://mcp-falaai.action.tec.br/mcp: tools/list returned 5 tools to an anonymous client, including risk_audit (write), transcribe_audio (write), extract_whatsapp (write), health (write)
Findings
Matches of fixed checks, with the text they matched.
- Inforisk_audit
The description is 2,052 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ning), and persist it if the user may need it later; calling again charges again."
- Infodiagnostic
The description is 2,008 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ning), and persist it if the user may need it later; calling again charges again."
- Infoextract_whatsapp
The description is 2,909 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ning), and persist it if the user may need it later; calling again charges again."
History
Changes to its tool definitions and to its remote endpoint, newest first.
Tools2 changed.~2 changedView diff
- Changed:extract_whatsappChanges thingsdescription
- Changed:transcribe_audioChanges thingsdescription
Install
Pinned to version 1.21.67. Values in angle brackets are yours to fill in.
claude mcp add --transport http falaai https://mcp-falaai.action.tec.br/mcpclaude mcp add --transport http falaai https://mcp-falaai.action.tec.br/mcpRun in a terminal.
Claude Code permissions
Allow the reads, ask before a change or a way out, deny the deletes. Paste into .claude/settings.json.
{
"permissions": {
"allow": [
"mcp__falaai__diagnostic"
],
"ask": [
"mcp__falaai__risk_audit",
"mcp__falaai__transcribe_audio",
"mcp__falaai__extract_whatsapp",
"mcp__falaai__health"
],
"deny": []
}
}Tools that only read are allowed, tools that change things ask first, tools that delete or cancel are denied. A read that can send data out (it takes a URL, say) asks first too. It assumes the server is installed as falaai. Review it before you use it.
Capability tags behind 1 rule
- mcp__falaai__diagnostic private-data
Pin
Keep the decision: fail CI when the tools change, follow every change, or show a badge.
Pin in CI
Fail a build when the tools you reviewed are not the tools the server serves today.
curl -s "https://radar.anlyon.com/api/v1/servers/io.github.ActionTecBr/falaai-mcp/verify?hash=cf724e81c7f1029b129a350f752df14d90fc562a005f0259df07c3b33d6529c8" | jq -e '.changed == false'Pinned hash cf724e81c7f1, from Nov 16, 2026, 16:17 UTC.
Watch
Get told when its tools change: a new tool, a new description, a new class.
https://radar.anlyon.com/servers/io.github.ActionTecBr/falaai-mcp/changes.xmlAn Atom feed. Works in any feed reader or Slack's RSS app.
Badge
[](https://radar.anlyon.com/servers/io.github.ActionTecBr/falaai-mcp)Source: MCP Registry. Synced Oct 21, 2026, 18:14 UTC.
FalaAI API (MCP): questions and answers
Answered from the scan above, not from the publisher's description.
Is the FalaAI API (MCP) safe?
- No scan can promise that. What its definitions show: 5 tools, 4 of which can change, send or delete something, and 3 findings from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.
Can the FalaAI API (MCP) delete data?
- No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.
Can the FalaAI API (MCP) do things that cannot be undone?
- No tool is classed as irreversible: none of them sends, pays, publishes or merges, going by their definitions.
What tools does the FalaAI API (MCP) have?
- 5 tools: 4 write, 1 read. risk_audit, transcribe_audio, extract_whatsapp, health and diagnostic.
Does the FalaAI API (MCP) need sign-in or OAuth?
- Its remote endpoint listed its tools to an anonymous client, without sign-in.
Does the FalaAI API (MCP) send data to outside services?
- None of its tools sends, posts or publishes, or takes a recipient, channel or URL.
Has the FalaAI API (MCP) changed its tools?
- Yes, once since Anlyon Radar started tracking it, most recently on Nov 3, 2026. Every change is on this page and in its Atom feed.