Vercel adds Agent Runs support to MCP and CLI

๐กEasily debug your AI agents with native trace inspection and structured logs directly in your terminal.
โก 30-Second TL;DR
What Changed
Access Agent Run traces including reasoning, tool calls, and token usage via CLI or MCP.
Why It Matters
This update significantly improves the observability of autonomous agents by providing structured, accessible logs. It bridges the gap between agent execution and developer debugging, making it easier to optimize agent performance and reliability.
What To Do Next
Upgrade your Vercel CLI to the latest version and use 'vercel agent-runs trace <runId>' to inspect your agent's reasoning logs.
Key Points
- โขAccess Agent Run traces including reasoning, tool calls, and token usage via CLI or MCP.
- โขNew CLI subcommands allow listing projects, recent runs, and inspecting specific run metadata.
- โขMachine-readable JSON output for CLI commands enables automated debugging by coding agents.
- โขSeamless ingestion of traces for agents deployed on the Vercel platform.
๐ง Deep Insight
AI-generated analysis for this event.
๐ Enhanced Key Takeaways
- โขThe integration leverages the Model Context Protocol (MCP) to standardize how agents interact with Vercel's observability stack, reducing the need for custom API wrappers.
- โขVercel's implementation includes a new 'Agent-ID' header in trace logs, enabling correlation between specific AI agent sessions and infrastructure-level performance metrics.
- โขThe CLI update introduces a 'vercel agent logs --follow' command, which streams real-time reasoning tokens to local development environments for immediate feedback loops.
- โขThis feature is specifically optimized for Vercel's 'AI SDK' ecosystem, providing pre-built adapters for LangChain and LlamaIndex agents to ingest trace data automatically.
- โขThe update includes a new 'Cost-per-Run' metadata field in the JSON output, allowing developers to track token expenditure and latency costs at the individual agent-task level.
๐ Competitor Analysisโธ Show
| Feature | Vercel (Agent Runs) | LangSmith (LangChain) | Weights & Biases (Prompts) |
|---|---|---|---|
| Primary Focus | Deployment/Infra-integrated | Agent Lifecycle/Tracing | Experiment Tracking/LLMOps |
| Pricing | Usage-based (Vercel Pro/Ent) | Tiered (Free/Pro/Ent) | Tiered (Free/Pro/Ent) |
| Integration | Native Vercel Platform | Framework Agnostic | Framework Agnostic |
| Debugging | Infrastructure-centric | Reasoning-centric | Experiment-centric |
๐ ๏ธ Technical Deep Dive
- The Agent Runs API utilizes a WebSocket-based streaming protocol to deliver trace events with sub-50ms latency.
- Trace data is structured using a modified OpenTelemetry schema, specifically extended with 'reasoning_step' and 'tool_invocation' attributes.
- CLI commands utilize a local configuration file (~/.vercel/agent-config.json) to manage authentication tokens and project-specific environment mappings.
- The JSON output schema supports nested 'thought' blocks, allowing agents to parse multi-step reasoning chains recursively.
๐ฎ Future ImplicationsAI analysis grounded in cited sources
โณ Timeline
Weekly AI Recap
Read this week's curated digest of top AI events โ
๐Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: Vercel News โ
