Livepeer Frameworks Monorepo Skills

FrameWorks

Multi-tenant live streaming platform with three access layers and crypto-native auth.

Skill Files

Platform Overview

Interfaces

Authentication Methods

What You Can Do

MCP: 29 tools, 18 resources, 8 prompts — full discovery via tools/list and resources/list. GraphQL: introspection enabled at /graphql — full schema discovery built-in.

Security Notes

• Never share private keys or seed phrases with third parties.
• Store agent credentials locally (see Credentials below).
• Only send authentication headers to *.frameworks.network domains.

Credentials

Store credentials at ~/.config/frameworks/credentials.json:

{
  "wallet_address": "0x...",
  "jwt": "eyJ...",
  "api_base": "https://bridge.frameworks.network"
}

Or use environment variables: FRAMEWORKS_WALLET_PRIVKEY, FRAMEWORKS_JWT.

Quick Start (Agent Flow)

1. Create or load an EVM wallet.
2. Sign a wallet login message (EIP-191) and call POST /auth/wallet-login to auto-provision a prepaid tenant.
3. Fund the tenant via x402 or crypto deposit.
4. Connect via MCP (POST /mcp) or GraphQL (POST /graphql) with wallet headers or JWT.
5. Create a stream and push RTMP using the stream key.

Wallet Authentication

Headers:

• X-Wallet-Address: 0x...
• X-Wallet-Signature: 0x... (EIP-191 personal_sign)
• X-Wallet-Message: <exact message>

Message format (verbatim):

FrameWorks Login
Timestamp: 2025-01-15T12:00:00Z
Nonce: 12345

Wallet login endpoint: POST /auth/wallet-login

MCP Configuration

Discovery: GET /.well-known/mcp.json Endpoint: POST /mcp Transport: HTTP + SSE (streamable-http)

Example (Claude Desktop)

{
  "mcpServers": {
    "frameworks": {
      "url": "https://bridge.frameworks.network/mcp",
      "headers": {
        "X-Wallet-Address": "0x...",
        "X-Wallet-Signature": "0x...",
        "X-Wallet-Message": "FrameWorks Login\nTimestamp: 2025-01-15T12:00:00Z\nNonce: 12345"
      }
    }
  }
}

x402 Payments

Gasless USDC payments for instant top-ups or per-request auth. Also acts as an economic barrier against automated abuse.

• Header: X-PAYMENT: <base64 payload>
• Supported networks: Base, Arbitrum (USDC)

GraphQL (Alternative Interface)

Endpoint: POST /graphql

Key operations:

• Mutations: createStream, updateStream, deleteStream, refreshStreamKey
• Queries: streams, stream, me, balance
• Subscriptions: streamHealthUpdated

Authentication: same wallet headers or bearer token.

Rate Limits & Billing

• API requests are free; usage costs apply to viewer hours, storage, and processing.
• Prepaid balance must be positive to run billable operations.
• Use MCP billing://balance or GraphQL balance query to monitor balance and drain rate.

Streaming Best Practices

• Check balance before creating streams. Active streams drain balance continuously. Use billing://balance (MCP) or balance query (GraphQL) to check drain rate.
• Monitor stream health. Read streams://{id}/health periodically during live streams. Use QoE diagnostic tools if viewers report issues.
• Top up proactively. Streams are terminated if balance drops below -$10. Use x402 auto-payment or topup_balance to maintain buffer.
• Clean up after yourself. Delete streams, clips, and VOD assets you no longer need. Storage costs are ongoing.

Video Consultant (Skipper)

Use ask_consultant to query the Skipper pipeline — knowledge retrieval, query rewriting, reranking, optional web search, and multi-step reasoning. Every answer includes confidence tagging and source citations.

Knowledge Domains

Read knowledge://sources for the live list of indexed URLs and sitemaps.

Effective Queries

• Be specific: include protocol, codec, or tool name
  • Good: "How do I configure SRT latency in MistServer for a 500ms target?"
  • Weak: "How to reduce latency?"
• Platform questions: mention "FrameWorks" explicitly to prioritize platform docs
• Mode: "docs" for factual lookups (faster, no web), "full" (default) for web-augmented reasoning
• Iterate: if confidence is best_guess or unknown, rephrase with different terminology

Confidence Tags

Tool Composition

For stream diagnostics, collect data first, then interpret:

1. get_stream_health_summary — overview (bitrate, FPS, issues)
2. Symptom-specific tool (diagnose_rebuffering, diagnose_buffer_health, diagnose_packet_loss, diagnose_routing)
3. ask_consultant — pass diagnostic JSON for expert recommendations

Guided Workflows

Node Management

Agents that provision their own edge infrastructure can manage node lifecycle:

1. Create cluster: create_private_cluster — get bootstrap enrollment token
2. Add nodes: create_enrollment_token for additional edges in the same cluster
3. Provision: frameworks edge provision --enrollment-token <token> --ssh user@host
4. Check health: get_node_info for registration data, get_node_health for live metrics (CPU, RAM, bandwidth, active viewers)
5. Set mode via API: set_node_mode — no SSH needed, goes through Gateway → Commodore → Foghorn
6. Set mode via CLI: frameworks edge mode draining / maintenance / normal (local or --ssh user@host)
7. Diagnose: frameworks edge doctor + frameworks edge logs

Two management paths: use set_node_mode / get_node_health MCP tools when you don't have SSH access. Use CLI commands when you're on the edge or have SSH. Use manage_node for guided CLI command generation.

When to Alert Your Human

Do alert:

• Balance is critically low (< $5 with active streams)
• Stream health shows critical status
• Billing details are missing and you can't proceed
• x402 payment settlement fails
• Wallet signature is rejected (may need re-signing)

Heartbeat (Periodic Check)

Run every 15–30 minutes during active streaming, every few hours otherwise.

1. Account health: Read account://status. Resolve any blockers.
2. Balance: Read billing://balance. Alert human if < $5 with active streams.
3. Active streams: Read streams://list. For each live stream, read streams://{id}/health. If status: critical, run diagnose_rebuffering and diagnose_buffer_health.
4. Skill updates: Check skill.json version periodically (once/day).

If nothing notable: no output needed. If action required: surface the specific issue and recommended resolution.

For the full periodic check routine, load heartbeat.md.

Preflight Errors

Billable MCP tools run preflight checks before execution. These are the blocking errors:

Rate limiting is handled at the Gateway layer (HTTP 429) with standard Retry-After headers — not as a preflight error. Free operations (reads, listing, health checks) skip preflight entirely.

Example: First Stream

1. Authenticate — Sign EIP-191 message ("FrameWorks Login\nTimestamp: <ISO8601>\nNonce: <random>"), call POST /auth/wallet-login → receive JWT + tenant auto-provisioned.
2. Connect — POST /mcp or POST /graphql with Authorization: Bearer <jwt> or wallet headers.
3. Resolve blockers — Read account://status → check blockers. Fix BILLING_DETAILS_MISSING with update_billing_details, INSUFFICIENT_BALANCE with x402 or topup_balance.
4. Create & stream — create_stream → capture stream_key + rtmp_url. Push RTMP: rtmp://<ingest>/live/<stream_key>.
5. Monitor — Read streams://{id}/health periodically. If issues: diagnose_rebuffering, diagnose_buffer_health.
6. Wrap up — delete_stream or leave. Check billing://balance for cost.

Converted and distributed by TomeVault — claim your Tome and manage your conversions.