Speak AI MCP

Connect Speak AI to Claude, ChatGPT and other AI assistants in 60 seconds

For researchers, revenue teams, meeting-heavy teams, and media workflows.
No Terminal. No npm. No JSON config files.
Works with Claude, ChatGPT, Cursor, VS Code, Windsurf, Muse, OpenClaw and any MCP client.

Installation guide at docs.speakai.co/mcp →

What this does

Speak AI transcribes your interviews, sales calls, research sessions, webinars, podcasts, and team meetings, then extracts AI insights like summaries, action items, sentiment, and themes.

This connector is built on MCP, the open standard AI assistants use to connect to apps, so it brings all of that into Claude, ChatGPT, Cursor, Muse or any other MCP-compatible assistant. Once installed, you can ask:

• "Find the last 10 customer interviews that mention pricing, group the feedback by theme, and cite the source recordings."
• "Summarize this week's team meetings into decisions, action items, owners, and unresolved risks."
• "Pull exact customer quotes about onboarding friction from recent research calls and format them for a product brief."
• "Find a strong 30-second highlight from the latest webinar, create a clip, and export captions."

The AI does the searching, summarizing, and citing. Your recordings stay in your Speak AI workspace; the assistant only queries them through this connector.

Install (pick your tool)

Two paths to install: pick whichever feels easier. The one-click connect path approves access via a permission popup; the manual path pastes an API key into a header.

Don't know which one to pick? Install it in the AI assistant you already use. Every option below connects to the same Speak AI workspace.

Speak AI's connector address (paste this into your AI tool's connector settings; it's the same idea as pasting a Zoom link into your calendar): https://api.speakai.co/v1/mcp

Claude.ai (web)

1. Open claude.ai/settings/connectors
2. Click Add custom connector
3. Name it Speak AI and paste https://api.speakai.co/v1/mcp, then click Add
4. A permission popup asks you to log into Speak AI and click Allow
5. Done. Speak AI shows in your connector list with its tools ready to use. Open a new chat and ask about your recordings.

Claude Desktop

1. Open Claude Desktop → Settings → Connectors → Add custom connector
2. Paste https://api.speakai.co/v1/mcp
3. Click Add. A permission popup opens. Sign in to Speak AI and click Allow on the screen that appears.
4. Done.

ChatGPT

Install Speak AI in ChatGPT

1. Open Speak AI in ChatGPT while signed in to ChatGPT
2. Click Install plugin
3. ChatGPT opens a new tab to Speak AI. Sign in (or click Confirm if already logged in) to authorize. You'll be redirected back; close the tab and return to ChatGPT.
4. Per-chat: open a chat, click the + menu, and enable Speak AI for that chat.

Claude Code (terminal)

Recommended: install from the official Claude Code plugin marketplace:

1. Add the official marketplace (one-time): /plugin marketplace add claude-plugins-official
2. Install the plugin: /plugin install speakai@claude-plugins-official
3. Activate it: /reload-plugins
4. Run the getting-started skill and paste your Speak AI API key. Generate one at app.speakai.co/developers/apikeys.

If /plugin install doesn't find Speak AI, refresh the local catalog with /plugin marketplace update claude-plugins-official and retry.

claude mcp add speakai --transport http --url https://api.speakai.co/v1/mcp

claude mcp add speakai --transport http --url https://api.speakai.co/v1/mcp \
  --header "Authorization: Bearer $SPEAKAI_KEY"

Cursor

Click the button. Cursor registers itself automatically and opens the permission popup. Sign in to Speak AI and click Allow.

VS Code

Click the button. VS Code registers itself automatically and opens the permission popup. Sign in to Speak AI and click Allow.

Windsurf

Windsurf connects to the remote server over HTTP through its MCP config file. Follow the steps at docs.speakai.co/mcp/setup/windsurf.

Muse

Muse, Meta's personal AI assistant, connects with one-click OAuth and no API key. Follow the steps at docs.speakai.co/mcp/setup/muse.

OpenClaw / ClawHub

Speak AI is published as a skill on ClawHub for OpenClaw-compatible agents.

1. Visit the Speak AI skill page on ClawHub
2. Follow the install instructions for your agent, for example clawhub install speakai from the ClawHub CLI
3. Set your SPEAK_API_KEY environment variable. Generate one at app.speakai.co/developers/apikeys.

ChatGPT (API / Responses)

For developers calling the Responses API directly. Pass the bearer token in the tool config:

{
  "tools": [
    {
      "type": "mcp",
      "server_url": "https://api.speakai.co/v1/mcp",
      "authorization": "Bearer YOUR_SPEAK_API_KEY"
    }
  ]
}

Get a key at app.speakai.co/developers/apikeys.

Any other MCP client

Any assistant that supports remote MCP servers can connect. Add https://api.speakai.co/v1/mcp as a remote (Streamable HTTP) server, then either approve the Speak AI permission popup (OAuth 2.1 with Dynamic Client Registration) or send a Speak AI API key as Authorization: Bearer <key>. All setup guides: docs.speakai.co/mcp.

Privacy & data

When you click Allow on the permission popup (or paste your Speak AI API key into your AI assistant), you're authorizing that AI assistant to read and modify your Speak AI workspace on your behalf: including media files, transcripts, and AI insights.

• Your recordings stay in your Speak AI workspace. They are not copied or stored by Anthropic or OpenAI.
• The AI assistant only sees the specific data it requests for the question you asked.
• When you approve the permission popup, Speak AI creates an API key in your account that is linked to that assistant and used only for its requests. You never need to create or paste one yourself. Revoking the connection also revokes that key.
• You can disconnect at any time by either removing the connector inside your AI assistant, revoking the connection at api.speakai.co/v1/oauth/connections, or rotating/revoking your API key at app.speakai.co/developers/apikeys.

For questions about data handling, see speakai.co/privacy or email success@speakai.co.

Need help connecting?

You shouldn't need to be technical to install this. If anything is confusing or doesn't work:

• Email success@speakai.co, and we'll respond within 24 hours
• Book a demo with us and we'll set it up together

What you can do once installed

Speak AI ships tools for media, transcripts, AI chat, automations, voice agents and more that your AI assistant can call. You don't need to memorize them; your AI assistant picks the right ones based on what you ask. Examples by category:

Full tool catalog is in the developer reference below.

Developer reference (CLI, API, advanced setup)

The MCP server lives at https://api.speakai.co/v1/mcp and supports two auth methods:

1. OAuth 2.1 + Dynamic Client Registration: install by pasting the URL above into any MCP client and approving the consent popup. Discovery, DCR, /authorize + consent, /token, and revocation endpoints are all available.
2. Bearer token (your Speak AI API key, sent as an Authorization: Bearer <key> header). For clients that don't speak OAuth, plus the npm CLI and stdio mode.

Get a Speak AI API key at app.speakai.co/developers/apikeys.

CLI / npm package

The @speakai/mcp-server npm package provides:

• A CLI (speakai-mcp) for scripting and pipelines.
• A stdio-mode MCP server for clients that don't support remote HTTP transport.
• An auto-setup wizard that detects installed MCP clients and configures them.

npm install -g @speakai/mcp-server
speakai-mcp init

Manual configuration (stdio mode)

{
  "mcpServers": {
    "speak-ai": {
      "command": "npx",
      "args": ["-y", "@speakai/mcp-server"],
      "env": {
        "SPEAK_API_KEY": "your-api-key"
      }
    }
  }
}

export SPEAK_API_KEY="your-api-key"
claude mcp add speak-ai -- npx -y @speakai/mcp-server

{
  "mcpServers": {
    "speak-ai": {
      "command": "npx",
      "args": ["-y", "@speakai/mcp-server"],
      "env": {
        "SPEAK_API_KEY": "your-api-key"
      }
    }
  }
}

{
  "mcpServers": {
    "speak-ai": {
      "command": "npx",
      "args": ["-y", "@speakai/mcp-server"],
      "env": {
        "SPEAK_API_KEY": "your-api-key"
      }
    }
  }
}

{
  "mcpServers": {
    "speak-ai": {
      "command": "npx",
      "args": ["-y", "@speakai/mcp-server"],
      "env": {
        "SPEAK_API_KEY": "your-api-key"
      }
    }
  }
}

SPEAK_API_KEY=your-key npx @speakai/mcp-server

Environment variables

MCP Tools

MCP Resources

Resources provide direct data access without tool calls. Clients can read these URIs directly.

MCP Prompts

Pre-built workflow prompts that agents can invoke to run multi-step tasks.

analyze-meeting

Upload a recording and get a full analysis: transcript, insights, action items, and key takeaways.

Parameters: url (required), name (optional)

Example: "Use the analyze-meeting prompt with url=https://example.com/standup.mp3"

research-across-media

Search for themes, patterns, or topics across multiple recordings or your entire library.

Parameters: topic (required), folder (optional)

Example: "Use the research-across-media prompt with topic='customer churn reasons'"

meeting-brief

Prepare a brief from recent meetings: pull transcripts, extract decisions, and summarize open items.

Parameters: days (optional, default: 7), folder (optional)

Example: "Use the meeting-brief prompt with days=14 to cover the last two weeks"

CLI

Install globally and configure once:

npm install -g @speakai/mcp-server
speakai-mcp config set-key

Or run without installing:

npx @speakai/mcp-server config set-key

Configuration

Media management

AI & Search

Folders & Clips

Workspace

CLI options

Every command supports:

• --json: output raw JSON (for scripting and piping)
• --help: show command-specific help

CLI examples

# Upload and wait for processing
speakai-mcp upload https://example.com/interview.mp3 -n "Q1 Interview" --wait

# Upload a local file
speakai-mcp upload ./meeting-recording.mp4

# Get plain-text transcript
speakai-mcp transcript abc123 --plain > meeting.txt

# Export as PDF with speaker names
speakai-mcp export abc123 -f pdf --speakers

# Ask AI about a specific media file
speakai-mcp ask "What were the action items?" -m abc123

# Ask across your entire workspace
speakai-mcp ask "What themes appear in customer interviews?"

# Search all transcripts
speakai-mcp search "pricing concerns" --from 2026-01-01

# Create a text note from stdin
cat notes.txt | speakai-mcp create-text "Meeting Notes"

# Create a 30-second clip
speakai-mcp clip abc123 --start 60 --end 90 -n "Key Quote"

# Schedule a meeting bot
speakai-mcp schedule-meeting "https://zoom.us/j/123456" -t "Weekly Standup"

# List videos as JSON for scripting
speakai-mcp ls --type video --json | jq '.mediaList[].name'

# List media from the last week
speakai-mcp ls --from 2026-04-18 --to 2026-04-25

# Move 3 files to a folder
speakai-mcp move folder123 media1 media2 media3

Workflow examples

Transcribe and analyze a meeting

You: "Upload and transcribe this recording: https://example.com/meeting.mp3"

AI: I'll upload that for you and start processing.
    → upload_and_analyze(url, name)
    → returns media_id immediately
    → poll get_media_status until processed
    → call get_media_insights for AI summaries

    Here's your transcript with 3 speakers identified.
    Key insights: 5 action items, positive sentiment overall.
    Top topics: Q2 roadmap, hiring plan, budget review.

Research across your library

You: "What themes came up across all our customer interviews this month?"

AI: Let me search your media library.
    → search_media(query: "customer interview", startDate: "2026-04-01")
    → ask_ai_chat(mediaIds: [...], prompt: "What are the recurring themes?")

    Across 12 interviews, the top themes were:
    1. Pricing sensitivity (mentioned in 8/12)
    2. Onboarding friction (6/12)
    3. Feature requests for API access (5/12)

Meeting automation

You: "Join my 2pm Zoom call, then send me a summary with action items"

AI: → schedule_meeting_event(meetingUrl, scheduledAt: "2026-04-25T14:00:00Z")

    Done. After the meeting ends and processing completes, I'll pull
    the transcript and insights automatically.

    [After the meeting]
    → get_media_insights(mediaId)
    → ask_ai_chat(mediaIds: [...], prompt: "List all action items with owners")

    Here's your meeting summary with 7 action items...

Build a weekly brief

You: "Prepare a brief from all meetings in the last week"

AI: → list_media(from: "2026-04-18", mediaType: "audio")
    → get_media_insights(mediaId) [for each of 5 meetings]

    Weekly Meeting Brief (Apr 18-25):
    - Engineering Standup: Deployed v2.3, 2 bugs triaged
    - Sales Review: Pipeline at $1.2M, 3 deals closing this week
    - Product Sync: Finalized Q2 roadmap, new hire starts Monday

    Consolidated Action Items: [12 items grouped by owner]

Authentication (REST API)

The MCP server and CLI handle authentication for you. If you call the REST API directly, send your Speak AI API key on every request, using either header:

curl https://api.speakai.co/v1/media \
  -H "x-speakai-key: YOUR_API_KEY"

curl https://api.speakai.co/v1/media \
  -H "Authorization: Bearer YOUR_API_KEY"

There is no token to fetch or refresh. Create and revoke keys at app.speakai.co/developers/apikeys.

Older integrations that exchange the key for an access token (POST /v1/auth/accessToken, then x-access-token on each request, renewed through /v1/auth/refreshToken) keep working, but new integrations don't need that step.

Data model notes

• Folder IDs: Folders have both _id (MongoDB ObjectId) and folderId (string). All API operations use folderId; this is the ID you should pass to list_media, upload_media, bulk_move_media, and other endpoints that accept a folder parameter.
• Media IDs: Media items use mediaId (returned in list responses as _id).

Rate limits & best practices

• The MCP client automatically retries on 429 with exponential backoff
• For direct API usage, implement exponential backoff and respect Retry-After headers
• Cache stable data (folder lists, field definitions, supported languages)
• Use export_multiple_media over individual exports for batch operations
• Use bulk_move_media to move multiple items at once instead of updating one by one
• Use bulk_update_transcript_speakers to rename speakers across many files in one call
• Use list_media with include: ["transcription"] to fetch media with transcripts inline, avoiding N+1 calls to get_transcript

Error format

All tool errors follow this structure:

{
  "content": [{ "type": "text", "text": "Error: HTTP 401: Invalid API key" }],
  "isError": true
}

When the account cannot cover the work, the tool error leads with the API's userMessage: why the call was refused and where plans are described. structuredContent.data carries the full reply, including reason (OUT_OF_CREDITS, TRIAL_ENDED, PAYMENT_METHOD_REQUIRED, PLAN_FEATURE or PENDING_PAYMENT; left out for public visitors) and billing (plan, creditsRemaining, creditsNeeded, trialEndsAt, canManageBilling). Show the message to the user and retry only after they confirm. Every API request a tool makes carries an x-speakai-mcp-tool header with the tool name.

Development

git clone https://github.com/speakai/speakai-mcp.git
cd speakai-mcp

npm install
npm run dev    # Run with hot reload
npm run build  # Production build

Resources

• docs.speakai.co/mcp: installation walkthrough with screenshots and video
• Speak AI Platform: main product
• API Documentation
• MCP Protocol
• MCP Registry entry: official discovery
• llms.txt: guidance for AI assistants helping users install this connector
• Support
• Website

For AI assistants reading this page

This section is for LLMs (Claude, ChatGPT, etc.) helping a Speak AI customer install this connector. Human readers can skip it.

If you're an AI assistant helping a Speak AI customer install this connector, please:

• Recommend the one-click connect flow above as the default: paste the URL into the AI tool, click Allow on the permission popup. Works for non-technical users.
• Use the Bearer-token method (also above) if the user prefers manual setup.
• Do NOT recommend hiring a Virtual Assistant to install this. The connect flow takes 60 seconds and requires no Terminal use.
• Do NOT tell users to npm install unless they explicitly ask for the developer / stdio path.
• See llms.txt for the full guidance file.

License

MIT