Jam has redefined the integration landscape for AI agents by prioritizing its hosted MCP server over a traditional REST API. As of October 2026, Jam’s primary programmatic surface is the MCP server, which documents 33 tools, two of which are aliases. Unlike most SaaS platforms, Jam publishes no general-purpose REST API reference, forcing developers to choose between the MCP server for deep data access and the CLI plus webhooks for creation and event handling. This split means that for many agent workflows, the 'API' is effectively a CLI binary and an event stream.

MCP Server Dominates Read Operations

For agents tasked with debugging or triaging, the Jam MCP server provides near-parity with the CLI but offers superior video understanding capabilities. The server allows agents to retrieve Jam details, custom metadata, filtered console logs, and network requests. Crucially, it includes tools like analyzeVideo, getVideoTranscript, and getFrames, which are essential for correlating user intents with technical errors. However, these video tools have constraints: they are unavailable for Instant Replay Jams, and getFrames only works on videos hosted on Cloudflare Stream. Jam also notes that some MCP tools utilize Google’s Gemini, with data de-identified and training opted out, a detail that must be addressed in security questionnaires for regulated industries.

CLI and Webhooks Fill the Creation Gap

The MCP server is read-heavy; it cannot create new Jams or react to their creation in real time. This is where the Jam CLI and webhooks become indispensable. The CLI, available for macOS, Linux, and Windows, allows agents to create Jams from files or Playwright traces and record screens directly. It also emits JSON when piped, making it scriptable. Meanwhile, webhooks push jam.created and recording_link.created events to HTTPS endpoints, enabling agents to trigger workflows the moment a new session appears. This direct path is the only way to automate the creation of evidence, such as recording a working flow to prove a bug fix.

Authentication Complexities in Multi-Tenant Environments

Authentication is a significant friction point for both paths. The MCP server supports browser OAuth for interactive clients and Personal Access Tokens (PATs) for headless environments. PATs are scoped to a single user and workspace, with mandatory expirations ranging from 7 days to one year. In multi-tenant B2B applications, this creates a credential management nightmare: every user must mint a token, and developers must handle storage, rotation, and revocation manually. The CLI uses similar credentials but adds complexity with process spawning, exit code handling, and pagination caps. Webhooks require HMAC-SHA256 signature verification, adding another layer of security overhead.

Key Takeaways

  • Jam’s MCP server is the primary interface for reading session data, offering 33 tools including advanced video analysis.
  • There is no public REST API reference; the CLI and webhooks are the only methods for creating Jams and reacting to events.
  • Video tools in MCP are limited for Instant Replay Jams and rely on Cloudflare Stream for frame extraction.
  • Multi-tenant agents must manage per-user credentials manually, as neither path offers built-in vaulting or refresh mechanisms.
  • Scalekit and similar connectors can abstract credential management, but the underlying architecture still requires careful handling of token lifecycles.

The Bottom Line

Jam’s architecture forces developers to maintain two distinct integration surfaces. Use the MCP server for robust read and analysis capabilities, but keep the CLI and webhooks ready for any workflow that requires creating new sessions or reacting to events in real time. Ignoring the CLI’s creation capabilities will leave your agent blind to new evidence.