Dalea AI

npx add-mcp 'https://app.usedalea.ai/api/mcp/mcp'

Dalea MCP server

Canonical: https://usedalea.ai/docs/mcp

The Dalea MCP server lets any Model Context Protocol client, such as Claude, ChatGPT, Claude Code or Cursor, work with your Dalea account: brands, content calendar, scripts, brainstorm, references, brand knowledge and Instagram and TikTok performance.

Every tool runs as the signed-in user and requires an active Dalea subscription. Without one, tools return an error that says so.

Server URL: https://app.usedalea.ai/api/mcp/mcp (Streamable HTTP; protocol versions 2025-11-25, 2025-06-18, 2025-03-26).

Connect

Claude

1. Open Claude (web or desktop) and go to Settings, then Connectors.
2. Choose Add custom connector and paste the server URL.
3. Click Connect, log in to Dalea and approve the access.

ChatGPT

1. Enable developer mode in ChatGPT settings.
2. Go to Connectors, create a new connector and paste the server URL.
3. Choose OAuth, save, then log in to Dalea and approve the access.

Claude Code

1. Add the server, then run /mcp inside Claude Code and choose Authenticate.

claude mcp add --transport http dalea https://app.usedalea.ai/api/mcp/mcp

Cursor and other clients

1. Create an API key in Dalea under Settings, MCP.
2. Add the server to your client's MCP configuration with the key as a Bearer token.

{
  "mcpServers": {
    "dalea": {
      "url": "https://app.usedalea.ai/api/mcp/mcp",
      "headers": {
        "Authorization": "Bearer dalea_mcp_..."
      }
    }
  }
}

Any MCP client

1. Any client that supports Streamable HTTP and OAuth 2.1 can connect with the server URL alone; it discovers the authorization server by itself.

From source to script

1. create_content_from_source: Start with a brand and editorial idea. Attach an article URL, 200 to 150000 characters of text, or a PDF uploaded with create_content_source_upload. Uploads are private, limited to 10 MiB and expire in two hours; send bytes to the signed URL without forwarding the Dalea bearer token. Use add_content_source for an existing card.

2. get_content_source: Poll extraction status until done. Read highlights and follow nextOffset for more text. Retry failed or stalled extractions with retry_content_source; remove unwanted sources with detach_content_reference. Each card supports five sources, including at most three PDFs.

3. attach_content_reference: Attach a catalog creator or video, or a video from the brand's connected accounts. Saved references expose videoId for attachment; savedVideoId identifies the library entry. Attaching a reference does not generate its analysis.

4. generate_scripts: Generate with the brand's voice, memory and extracted sources. set_content_web_search controls manual research; Dalea can also research time-sensitive topics automatically. Read citations with get_content or list_content_references. create_content_from_source with generate=true combines extraction and generation; on partial failure, resume with its returned card ID.

5. refine_content: Ask for an edit or discuss the script. By default Dalea saves a preview without changing the script or its settings. searchWeb=true researches the question and returns sources. get_refine_chat includes preview IDs, status and proposed changes.

6. review_content_refinement: Apply or dismiss the preview without spending another AI call. Applying refuses to overwrite a card changed since the preview. Optional settings proposals require includeSuggestedSettings=true. Dates and status only organize the editorial calendar; they never publish to Instagram or TikTok.

From performance to the next script

1. get_account_overview: Resolve the brand and check includeHealth=true. Accounts without snapshots are still connected and appear in accountsWithoutSnapshots. lastUpdatedAt uses sync or scrape freshness; an absent snapshot is not zero performance.

2. get_audience_growth: Read daily or weekly follower growth and views, observing partial periods, interpolated points and source. History currently follows the connection; do not promise preservation across reconnection.

3. get_performance_summary: Compare cumulative metrics of video posts published in each window. These are not interactions occurring during the window. deltaPct is relative percent change; deltaPercentagePoints.averageEngagementPct is the absolute change in percentage points. Use get_top_performers to select videos.

4. get_video: Read the full caption, saved analysis and transcript, nextStep, classificationSources and same-account comparison. Comparison engagement rates are ratios, while engagementRatePct is percent. A performance verdict requires at least five videos with views. Images and carousels have null view-based metrics; inspect metricAvailability.

5. analyze_video: Generate missing analysis of an own video using platformVideoId from list_videos and platform to disambiguate. Pass confirm=true for authorized analysis. Cached analysis is free; new analysis consumes the 10/day quota. Check saved and evidence: thumbnail or metadata is not observation of the video. Legacy language/evidence can be unknown. Manual classifications are preserved.

6. get_homepage_insights: Check freshness, currentVideoCount and scope before using stored insights. Reading never regenerates. generate_insights explicitly refreshes from up to 50 top-view videos in the last year, with at least ten eligible videos. videoCount is the total, sampleSize is the analyzed sample. Freshness follows language, format and count changes of five, not every metric update or age. Treat diagnoses as hypotheses, not measured retention or causal proof.

7. create_content: Turn agreed lessons into editorial ideas, then use generate_scripts and refine_content with the brand's strategy and references. The weekly-review prompt guides this workflow. This does not publish or schedule a social post.

Pages

• Authentication
• Rate limits & quotas
• Errors
• Prompts
• Tools