Danube — Connect Your Agent

Danube gives your AI agent access to 100+ services and 30 tools through a single API key.

Quick Setup

Step 1: Get an API Key

You can get your API key from the Danube Dashboard under Settings > API Keys.

Alternatively, use the standard OAuth 2.0 Device Authorization flow (RFC 8628) — this requires the user to explicitly approve access in their browser:

curl -s -X POST https://api.danubeai.com/v1/auth/device/code 
  -H "Content-Type: application/json" 
  -d '{"client_name": "My Agent"}'

This returns a device_code, a user_code, and a verification_url.

The user must open the verification URL in their browser and enter the code to authorize access.

Then poll for the API key:

curl -s -X POST https://api.danubeai.com/v1/auth/device/token 
  -H "Content-Type: application/json" 
  -d '{"device_code": "DEVICE_CODE_FROM_STEP_1"}'

428 = user hasn't authorized yet (keep polling every 5 seconds)
200 = success, response contains your api_key
410 = expired, start over

Step 2: Connect via MCP

Add this to your MCP config:

{
  "mcpServers": {
    "danube": {
      "url": "https://mcp.danubeai.com/mcp",
      "headers": {
        "danube-api-key": "YOUR_API_KEY"
      }
    }
  }
}

Security & Privacy

Credential encryption: All stored credentials are encrypted at rest using AES-256 with a dedicated encryption key. Credentials are never logged or exposed in tool responses.
User-scoped access: Each API key is scoped to the authenticated user. You cannot access another user's credentials, tools, skills, or wallet.
Row-level security: Database access is enforced with row-level security policies — queries only return data belonging to the authenticated user.
Spending controls: Wallet operations are protected by configurable per-call and daily spending limits (in USDC). Agents cannot exceed these caps.
Audit trail: All tool executions are logged with timestamps, parameters, and results for user review.
Credential retention: Stored credentials persist until the user explicitly deletes them via the dashboard. They are not shared with other users or services.

Permissions & Scope

The DANUBE_API_KEY grants: - Read: Browse services, search tools, view public skills/workflows/sites, check wallet balance and spending limits - Execute: Run tools and workflows (subject to spending limits) - Write (user-scoped only): Store credentials for your account, create/update/delete your own skills and workflows, manage your spending limits - Financial: Fund your own wallet (requires user confirmation), execute paid tools (within spending caps)

The API key does not grant: - Access to other users' data, credentials, or resources - Admin or platform-level operations - Ability to bypass spending limits - Access to raw database or infrastructure

Step 3: Use Tools

Once connected, you have access to 30 MCP tools:

Discovery - list_services(query, limit) — Browse available tool providers - search_tools(query, service_id, limit) — Find tools by what you want to do (semantic search) - get_service_tools(service_id, limit) — Get all tools for a specific service

Execution - execute_tool(tool_id, tool_name, parameters) — Call a specific, registered service integration (e.g. send an email, create a ticket). Each tool has a fixed schema — this is not arbitrary code execution. - batch_execute_tools(calls) — Call up to 10 registered service integrations concurrently in one request

Credentials & Wallet - store_credential(service_id, credential_type, credential_value) — Save a user-provided API key for a specific service. The agent must never solicit, infer, or guess credentials. Only store keys the user explicitly provides, and always confirm before storing. Credentials are encrypted at rest (AES-256) and scoped to the user's account. - get_wallet_balance() — Check your credit balance before running paid tools - get_spending_limits() — View your USDC per-call and daily spending limits - update_spending_limits(max_per_call_usdc, daily_limit_usdc) — Update your USDC spending limits. Requires user approval.

Skills - search_skills(query, limit) — Find reusable agent skills (instructions, scripts, templates) - get_skill(skill_id, skill_name) — Get full skill content by ID or name - create_skill(name, skill_md_content, ...) — Create a new private skill. Only operates on skills owned by the authenticated user. - update_skill(skill_id, name, skill_md_content, ...) — Update a skill you own - delete_skill(skill_id) — Delete a skill you own

Workflows - list_workflows(query, limit) — Browse public multi-tool workflows - create_workflow(name, steps, description, visibility, tags) — Create a new workflow. Only operates on workflows owned by the authenticated user. - update_workflow(workflow_id, name, description, steps, visibility, tags) — Update a workflow you own - delete_workflow(workflow_id) — Delete a workflow you own - execute_workflow(workflow_id, inputs) — Run a multi-tool workflow - get_workflow_execution(execution_id) — Check workflow execution results

Agent Web Directory - search_sites(query, category, limit) — Search the agent-friendly site directory - get_site_info(domain) — Get structured info about a website (pricing, docs, contact, FAQ, etc.)

Agent Management - register_agent(name, operator_email) — Register a new autonomous agent with API key and wallet - get_agent_info() — Get the current agent's profile and wallet balance - fund_agent_wallet(method, amount_cents) — Fund wallet via card checkout or USDC. Requires explicit user approval before funding.

Tool Quality - submit_rating(tool_id, rating, comment) — Rate a tool 1-5 stars - get_my_rating(tool_id) — Check your existing rating for a tool - get_tool_ratings(tool_id) — Get average rating and count for a tool - report_tool(tool_id, reason, description) — Report a broken or degraded tool - get_recommendations(tool_id, limit) — Get tool recommendations based on co-usage patterns

When a Tool Needs Credentials

If execute_tool returns an auth_required error, it means the service needs an API key. You must ask the user to provide the key — never generate, guess, or infer credentials. The user can also configure credentials directly at https://danubeai.com/dashboard instead. Once the user explicitly provides a key and confirms they want to store it, call:

store_credential(service_id="...", credential_type="bearer", credential_value="the_key")

Then retry the tool. Stored credentials are encrypted at rest and only accessible by the user's account.

Core Workflow

Every tool interaction follows this pattern:

Search — search_tools("what you want to do")
Check auth — If the tool needs credentials, ask the user for the key and use store_credential, or guide the user to https://danubeai.com/dashboard
Gather parameters — Ask the user for any missing required info
Execute — execute_tool(tool_id, parameters)
Report — Tell the user what happened with specifics, not just "Done"

tools-marketplace

Installation