How to Connect ScreenApp to OpenClaw AI Agents via MCP Server

Your AI agent should be able to pull up last Tuesday’s meeting notes without you digging through folders. With ScreenApp’s API and the Model Context Protocol (MCP), you can give OpenClaw agents direct access to your recordings, transcripts, and AI summaries.

This guide walks through building a ScreenApp MCP server skill so your OpenClaw agent can search recordings, fetch transcripts, and run AI analysis on your content, all from a chat message on Telegram, WhatsApp, or Discord.

If you are new to MCP servers for meeting recordings, check out our comparison of the best MCP servers for meetings to see how ScreenApp stacks up.

What is OpenClaw?

OpenClaw is a self-hosted gateway that connects messaging apps (WhatsApp, Telegram, Discord, iMessage) to AI coding agents. You run a single process on your machine and it bridges your chat apps to an always-available AI assistant.

The key idea: you message your agent from your phone, and it can use tools, read files, browse the web, and now, access your ScreenApp recordings.

OpenClaw uses skills to teach agents how to use tools. Each skill is a folder with a SKILL.md file that describes available commands. Skills can also wrap MCP servers, which is how we will connect ScreenApp.

OpenClaw discovers skills through a three-tier system. On session start, the agent scans the name and description from every skill’s frontmatter. When a skill looks relevant, it loads the full SKILL.md. This means your description field is the single most important factor for whether the agent picks your skill. Write it like you are telling a coworker what the skill does and when to use it.

What is MCP?

Model Context Protocol is a standard created by Anthropic that lets AI models access external data sources. Think of it like USB-C for AI apps. Instead of copy-pasting transcripts into ChatGPT, an MCP server lets the AI pull what it needs directly.

MCP servers run locally and expose tools (actions the AI can take) and resources (data the AI can read). For ScreenApp, the tools would include things like “search recordings,” “get transcript,” and “ask a question about a video.”

MCP works with Claude Desktop, ChatGPT Desktop, Cursor, and through OpenClaw, any messaging app.

ScreenApp API overview

ScreenApp provides a REST API that covers everything you need for an MCP integration. The main endpoints:

• List recordings - Browse your recording library with filters
• Get recording with transcript - Fetch a recording and its full transcript, with timestamps and speaker labels
• AI analysis - Ask questions about a recording using the multimodal analysis endpoint (/v2/files/{fileId}/ask/multimodal)
• Search - Find recordings by keyword across all your content
• Upload - Send new audio/video files for processing

Authentication uses Bearer tokens. You generate an API token from your ScreenApp account settings.

For the full reference, see the ScreenApp API documentation.

Setup steps

1. Get your ScreenApp API token

Log into ScreenApp, go to Settings > API, and generate a new token. Copy it somewhere safe. You will need it in the next step.

2. Create the skill folder

OpenClaw skills live in ~/.openclaw/skills/ (shared across agents) or in your workspace’s /skills/ folder (per-agent). The skill folder name must match the name field in your SKILL.md frontmatter. Create the full structure:

mkdir -p ~/.openclaw/skills/screenapp/references

The final folder layout:

screenapp/
├── SKILL.md              # Required: frontmatter + agent instructions
├── claw.json             # For ClawHub publishing (optional)
└── references/
    └── api-endpoints.md  # Detailed endpoint docs loaded on demand

3. Write the SKILL.md

The description field is what determines whether the agent picks this skill. OpenClaw scans every installed skill’s name and description at the start of each session. Include trigger phrases and keyword variants so the agent matches it to requests about transcripts, meetings, recordings, and note-taking.

Create ~/.openclaw/skills/screenapp/SKILL.md:

---
name: screenapp
description: Search meeting recordings and transcripts, get meeting notes and action items, transcribe audio and video, ask AI questions about recorded content. Use when the user mentions transcription, meeting notes, recordings, audio-to-text, speech-to-text, meeting search, or wants to find or analyze recorded meetings and calls.
metadata: {"clawdbot":{"emoji":"","requires":{"env":["SCREENAPP_API_TOKEN"]},"primaryEnv":"SCREENAPP_API_TOKEN","homepage":"https://screenapp.io"}}
---

# ScreenApp Integration

Access the user's ScreenApp recordings, transcripts, and AI analysis.

## Authentication

All requests require the SCREENAPP_API_TOKEN environment variable.

Headers: Authorization: Bearer $SCREENAPP_API_TOKEN
Base URL: https://api.screenapp.io

## Available actions

### Search recordings by keyword
GET /v2/files?searchQuery={query}

Use this when the user wants to find a specific meeting, recording, or topic.

### Get a recording with its transcript
GET /v2/files/{fileId}

Returns the full recording object including the timestamped transcript with speaker labels. Use this to retrieve meeting notes, read back what was said, or get the full text of a call.

### Ask AI a question about a recording
POST /v2/files/{fileId}/ask/multimodal
Body: { "promptText": "your question here" }

Analyzes both the transcript and video content. Use this for summarizing meetings, extracting action items, or answering specific questions about what happened in a recording.

### List recent recordings
GET /v2/files?sortBy=createdAt&sortOrder=desc&limit=10

Use this when the user asks about their latest meetings or recent recordings.

Optionally, create references/api-endpoints.md with more detailed parameter documentation. The agent loads reference files only when it needs them, keeping the initial skill scan lightweight.

4. Set the API token

Add your ScreenApp token to OpenClaw’s environment. Edit ~/.openclaw/openclaw.json:

{
  "env": {
    "SCREENAPP_API_TOKEN": "your-token-here"
  }
}

Or export it in your shell profile if you prefer.

5. Restart OpenClaw

openclaw gateway restart

The agent will pick up the new skill on the next session. You can verify it loaded by asking your agent: “What skills do you have?“

6. Test it

Send a message to your OpenClaw agent on Telegram or WhatsApp:

• “Search my ScreenApp recordings for the product roadmap discussion”
• “Get the transcript from my last meeting”
• “What action items came out of yesterday’s standup?”

The agent will use the ScreenApp skill to fetch the data and answer your question.

Publishing to ClawHub

If you want other OpenClaw users to find and install your ScreenApp skill, publish it to ClawHub, OpenClaw’s official skill registry. ClawHub uses vector search, so natural language queries like “search meeting transcripts” or “transcribe audio recordings” will surface your skill as long as the description covers those terms.

Add a claw.json manifest to your skill folder:

{
  "name": "screenapp",
  "version": "1.0.0",
  "description": "Search meeting recordings, get transcripts, extract action items, and ask AI questions about recorded content via the ScreenApp API.",
  "author": "your-clawhub-username",
  "license": "MIT",
  "permissions": ["network"],
  "entry": "SKILL.md",
  "tags": ["transcription", "meetings", "recordings", "note-taking", "productivity", "speech-to-text"],
  "models": ["claude-*", "gpt-*", "gemini-*"],
  "minOpenClawVersion": "0.8.0"
}

Then publish:

clawhub publish ~/.openclaw/skills/screenapp

Once published, anyone can install it with:

clawhub install screenapp

Your skill will also appear on Moltbook, a cross-platform index that lists skills for OpenClaw, Claude Code, Cursor, and other AI coding tools. Moltbook indexes ClawHub automatically, so there is no extra step.

Building an MCP server (advanced)

If you want a proper MCP server instead of a skill-based approach, you can wrap the ScreenApp API in a Node.js MCP server. This gives you compatibility with Claude Desktop, Cursor, and any other MCP client, not just OpenClaw.

// screenapp-mcp/index.js
import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { z } from "zod";

const server = new McpServer({
  name: "screenapp",
  version: "1.0.0",
});

const API_BASE = "https://api.screenapp.io";
const TOKEN = process.env.SCREENAPP_API_TOKEN;

server.tool(
  "search_recordings",
  { query: z.string() },
  async ({ query }) => {
    const res = await fetch(
      `${API_BASE}/v2/files?searchQuery=${encodeURIComponent(query)}`,
      { headers: { Authorization: `Bearer ${TOKEN}` } }
    );
    const data = await res.json();
    return {
      content: [{ type: "text", text: JSON.stringify(data, null, 2) }],
    };
  }
);

server.tool(
  "get_transcript",
  { fileId: z.string() },
  async ({ fileId }) => {
    const res = await fetch(`${API_BASE}/v2/files/${fileId}`, {
      headers: { Authorization: `Bearer ${TOKEN}` },
    });
    const data = await res.json();
    return {
      content: [{ type: "text", text: JSON.stringify(data.transcript, null, 2) }],
    };
  }
);

server.tool(
  "analyze_recording",
  { fileId: z.string(), question: z.string() },
  async ({ fileId, question }) => {
    const res = await fetch(
      `${API_BASE}/v2/files/${fileId}/ask/multimodal`,
      {
        method: "POST",
        headers: {
          Authorization: `Bearer ${TOKEN}`,
          "Content-Type": "application/json",
        },
        body: JSON.stringify({ promptText: question }),
      }
    );
    const data = await res.json();
    return {
      content: [{ type: "text", text: JSON.stringify(data, null, 2) }],
    };
  }
);

Then register it in OpenClaw’s config or use it standalone with Claude Desktop.

Use cases

Meeting follow-ups: Ask your agent “What did Sarah say about the timeline in yesterday’s call?” and it pulls the transcript, finds the relevant section, and gives you the answer.

Weekly summaries: “Summarize all my meetings from this week” triggers a search across your recordings and generates a consolidated report.

Action item tracking: “What open action items do I have from meetings this month?” searches your transcripts for commitments and deadlines.

Content repurposing: “Turn my last webinar recording into a blog post outline” uses the AI analysis endpoint to generate structured content from your video.

Related ScreenApp features

• AI Meeting Summarizer: Generate summaries from any recording
• Transcription Software: Automatic speech-to-text with speaker labels
• AI Note Taker: Structured notes from your conversations

FAQ

What is an MCP server?

An MCP (Model Context Protocol) server is a local program that exposes data and tools to AI models. It lets AI assistants like Claude or OpenClaw agents access external services without copy-pasting data manually.

Do I need a paid ScreenApp plan for API access?

API access is available on ScreenApp paid plans. Check your account settings to see if the API section is available. Free accounts have limited API access.

Can I use this with Claude Desktop instead of OpenClaw?

Yes. The MCP server approach works with any MCP-compatible client. Claude Desktop, ChatGPT Desktop, and Cursor all support MCP servers. OpenClaw adds the benefit of accessing it from mobile messaging apps.

Is my recording data sent to OpenClaw’s servers?

No. OpenClaw is self-hosted and runs entirely on your machine. Your ScreenApp API token and recording data stay local. The AI model processes the data, but OpenClaw itself does not store or forward your content.

How does the agent find the right skill?

OpenClaw scans the name and description fields from every installed skill at the start of each session. When you ask something like “find my last meeting,” the agent matches that against skill descriptions and loads the full SKILL.md for whichever skill fits best. This is why the description should include concrete trigger phrases like “transcription,” “meeting notes,” and “recordings” rather than vague language like “helps with recordings.”

How many recordings can the MCP server handle?

The ScreenApp API handles pagination, so you can work with libraries of any size. For best results, use search queries to narrow down results rather than listing everything at once.