DeepPane

Production MCP for AI agents

AI-callable ambient focus workspace and work-time dashboard.

DeepPane lets agents help a user create, share, and review focus spaces while the user's long-term work-time history remains private, authorized, compact, and revocable.

Start MCP quickstart Request Agent Key Read llms.txt Open Dashboard

Private intake

Use an Agent Key when the integration is real.

Registered Agent Keys identify the calling app for attribution, bounded quota, abuse controls, and agent provenance. They do not represent a user account, unlock Pro behavior, or expose private history.

Request Agent Key

Experimentation path

Anonymous creation is deliberately limited.

Anonymous calls are useful for first-touch demos and low-risk exploration, but they stay on Free policy, strict public limits, catalog quotes, no personal history attachment, and a 25-minute web music preview ceiling.

Agent Key request flow

Private review, then a scoped production key.

Agent Keys are issued through intake today. DeepPane shares the raw key once out-of-band, then keeps only a hashed ledger entry for attribution, quota, rotation, revoke, and abuse response.

  1. 01

    Submit partner and use-case details.

    Send product category, intended workflow, expected daily calls, user-authorized history needs, and target launch window.

  2. 02

    Review boundary and quota.

    DeepPane checks intended use, privacy boundary, media access boundary, abuse controls, and initial quota before issuing access.

  3. 03

    Receive the raw key once.

    The raw Agent Key is shared once through a private channel. The service stores only a hash and operational metadata.

  4. 04

    Set DEEPPANE_AGENT_KEY.

    Configure the key in the MCP host environment alongside DEEPPANE_API_BASE_URL=https://deeppane.com.

  5. 05

    Run smoke checks and operate.

    Run npm run check:mcp, verify Agent Key smoke, then watch alerts, rotation, and revoke readiness during pilot operations.

Production access

Official MCP is production-supported. OAuth and marketplace are separate tracks.

DeepPane's official stdio MCP path is available for production use with Dashboard handoff tokens. Registered Agent Keys are available by request for integrations that need attribution, bounded quota, abuse controls, and stable provenance.

Official MCP quickstart

Run the first-party server, then pass one short-lived handoff token.

This is the supported production path for MCP hosts today. It uses Dashboard consent, scoped tokens, and the no-media-library MCP boundary while OAuth and marketplace distribution remain separate future tracks.

1. Run MCP server

Start the first-party stdio server from your MCP host or local integration workspace.

2. Mint scoped token

Use Dashboard Agent Access to mint a short-lived focus-space or history-summary handoff token.

3. Call the tool

Call deeppane.create_focus_space or deeppane.read_history_summary with that token.

4. Hand off result

Return the short DeepPane link and PNG QR code to the user.

Start server

DEEPPANE_API_BASE_URL=https://deeppane.com node apps/mcp-server/server.js
  1. Open Dashboard Agent Access and choose either focus-space creation or history-summary access.
  2. Review the consent modal, create the token, then copy it from the MCP guide.
  3. Call the matching MCP tool with handoffToken. Do not paste a full DeepPane session token into an agent chat.

Tools

  • deeppane.create_focus_space
  • deeppane.request_focus_space_create_token
  • deeppane.request_history_summary_token
  • deeppane.read_history_summary

Production checks

npm run check:mcp
npm run check:agent:dogfood
npm run check:agent:staging-dogfood

The local checks verify MCP create/read/error behavior. The remote check repeats the path with Dashboard-minted short-lived tokens against the production API before release changes.

Forbidden boundary

The official server intentionally excludes media catalog browsing, raw history export, R2 keys, original files, signed media URLs, ranking weights, and bulk tag queries.

Positioning

Agent-friendly is the entry. Work-time history is the moat.

Agents can introduce a useful first focus experience, but DeepPane's durable value is the user's accumulated work-time history: calendar records, weekly and monthly trends, focus/ambient patterns, and agent-readable insights that become more valuable over time.

Create

Create or recommend a focus space for a task, recovery block, study sprint, writing session, or ambient work period.

Share

Hand the user a DeepPane link that can be opened, played, and measured without exposing the raw media library.

Track

Use the plugin and signed-in account to accumulate personal work-time history over days, weeks, and months.

Summarize

Read only user-authorized summaries that are safe for agent context and designed for concise review notes.

Why integrate

Give users a finished focus outcome, not another setup task.

DeepPane turns a user's intent into a playable workspace, share page, install path, and authorized history loop. Agent products get a reliable action to recommend while DeepPane keeps media, policy, and long-term work records inside the product boundary.

Outcome

Create a focus session from bounded intent hints without building a media pipeline or timer experience.

Conversion

Return share, install, and dashboard paths that can carry safe agent/run attribution into the product funnel.

Context

Let authorized users bring compact work-time summaries back to the agent for planning and review.

Trust

Keep session tokens, raw media, personal events, and catalog data out of agent prompts and wrappers.

Available now

Agent-created focus spaces

Agents can create one shareable, playable DeepPane focus space for a user task without receiving media files, signed media URLs, internal catalog scores, or bulk search results.

POST /v1/agent/focus-spaces
X-DeepPane-Agent-Key: <registered-agent-key> recommended
Authorization: Bearer <focus-space-create-token> optional
  • Anonymous/free creation supports bounded custom timers, catalog quotes, server-selected scenes, optional music, and strict public limits.
  • Registered agents should send X-DeepPane-Agent-Key for app attribution, bounded quota, and agent provenance; the key never represents a user's Pro entitlement.
  • User-authorized creation uses POST /v1/agent-auth/focus-space-create-token and applies the user's Free or Pro entitlement.
  • Agent-created web pages may preview music for up to 25 minutes. Longer uninterrupted music remains a plugin experience.
  • Intent, sceneHints, musicHints, quote intent, locale, and attribution.agentRunId are bounded hints, not media-library queries.
  • The response includes conversion.installUrl and conversion.dashboardUrl so the share page can lead to install, sign-in, and durable work-time history.

Available now

Agent-readable history summary

The stable endpoint for agents is intentionally narrow. It helps an agent understand the user's recent work rhythm without granting bulk export or raw event access.

GET /v1/focus-logs/history-summary?days=90
Authorization: Bearer <agent-handoff-token>
  • Returns audience: "agent" and purpose: "work_time_history_summary".
  • Includes metrics, period summaries, weekly trend, highlights, recommendations, and markdown.
  • Supports a compact days window from 7 to 180 days.

Scoped handoff

Separate tokens for separate powers.

Agents should never ask users for a full DeepPane session token. DeepPane issues short-lived handoff tokens for exactly one capability.

  • Mint creation tokens with POST /v1/agent-auth/focus-space-create-token.
  • Mint history tokens with POST /v1/agent-auth/history-summary-token.
  • focus_space.create can call POST /v1/agent/focus-spaces.
  • focus_logs.history_summary.read can call GET /v1/focus-logs/history-summary.
  • The two scopes are not interchangeable and do not grant account, billing, media-token, or raw export access.

MCP boundary

Expose callable abilities, not the library.

An MCP server or agent tool should wrap DeepPane's creation and history-summary endpoints with stable tool names. It should not expose a scene browser, media downloader, tag graph, ranking weights, R2 keys, or signed media URLs.

  • Recommended tools: deeppane.create_focus_space, deeppane.request_focus_space_create_token, deeppane.request_history_summary_token, and deeppane.read_history_summary.
  • Forbidden tools: list all scenes, list all music, download media, fetch original files, bulk-query tags, or reconstruct the catalog.
  • Forbidden wrapper names include deeppane.list_scenes, deeppane.list_music, deeppane.search_media_catalog, deeppane.get_signed_media_url, and deeppane.export_raw_focus_events.
  • Public discovery is descriptive only; access still requires endpoint policy, rate limits, and user authorization where applicable.

Example

What an agent should consume

Prefer the markdown field for short user-facing context. Use recommendations as suggestions, not commands.

{
  "schemaVersion": 1,
  "audience": "agent",
  "purpose": "work_time_history_summary",
  "window": { "days": 90, "fromDay": "2026-03-30", "toDay": "2026-06-27" },
  "metrics": {
    "totalFocusSeconds": 6900,
    "totalAmbientSeconds": 4200,
    "activeDays": 3,
    "currentStreakDays": 1
  },
  "highlights": [
    { "id": "overall-focus", "text": "Total focus time is 1h 55m across 3 active days." }
  ],
  "recommendations": [
    "Aim for three active focus days this week to strengthen the work rhythm."
  ],
  "markdown": "# DeepPane Work-Time History (90 days) ..."
}

Creation example

What an agent should request and return

Ask for the user's intent and let DeepPane choose matching media inside the product. Return the short link and PNG QR code; do not ask for raw files or a list of matching assets.

{
  "intent": "writing",
  "locale": "en",
  "timer": { "preset": "25_5" },
  "sceneHints": { "environment": ["forest"], "mood": ["calm"], "motion": "low" },
  "musicHints": { "use": ["focus"], "vocalProfile": "none", "energyLevel": "low" },
  "quote": { "mode": "catalog", "intent": "discipline" },
  "attribution": { "agentRunId": "run_123", "source": "agent" }
}
{
  "ok": true,
  "shareUrl": "https://deeppane.com/s/abc123XYZ?lang=en",
  "shortShareUrl": "https://deeppane.com/s/abc123XYZ?lang=en",
  "qrCodeUrl": "https://deeppane.com/s/abc123XYZ/qr.png?lang=en",
  "qrCodePngUrl": "https://deeppane.com/s/abc123XYZ/qr.png?lang=en",
  "qrCodeSvgUrl": "https://deeppane.com/s/abc123XYZ/qr.svg?lang=en",
  "policy": {
    "timerPreset": "25_5",
    "rawMediaAccess": false,
    "bulkMediaAccess": false
  },
  "selected": {
    "sceneTitle": "Forest Window",
    "musicTitle": "Quiet Piano Focus"
  }
}

For production integrations, include X-DeepPane-Agent-Key and a bounded agentRunId. DeepPane can then connect creation, share-page opening, server-side install/dashboard conversion attribution, and history attribution without logging secrets or raw user data. Run attribution is hash-bucketed before aggregate storage.

Plugin

The plugin creates durable history.

DeepPane's extension records focus clock completion, ambient work, session close or abandon, and related work-time events into the user's signed-in history ledger.

Account

The account unlocks long-term insight.

Registered users can carry history across sessions and let agents read compact summaries with authorization. Pro expands the human product experience; it does not grant agents bulk media access.

Roadmap boundary

Production MCP now. Hosted OAuth and marketplace later.

This page documents the implemented production MCP contract plus Agent Key intake. OAuth, agent marketplace registration, long-lived third-party authorization, and public agent pricing are separate product tracks and are not required for the current stdio MCP path.

Request Agent Key access