Monitors prompt-cache economics, displays restart pressure, and supports selective session handoffs through MCP.
Copy the AI prompt to install this server into Claude Code, Cursor, or another agent โ or use 1-click editor setup below.
We haven't yet run this listing's install command through our automated sandbox check. This isn't a red flag โ we're steadily working through the catalog.
๐ก Paste the JSON block into your client's configuration file under mcpServers, then restart the application.
Inspect callable tools, capabilities, and parameters exposed to AI agents by Session Watcher.
start_watcherStart (or reuse) the dashboard server; returns its URL
stop_watcherStop the managed server
watcher_statusReport whether the server is running and its URL
rotate_sessionRotate to a new session ID
get_bucket_summaryReturn current context bucket structure (files, skills, tools) with metrics
get_turn_skeletonRender the turns of the capture epoch, one block per turn, as the slots a note can fill
The nomadop/session-watcher MCP server treats prompt-cache tokens as session inventory. It follows a Claude Code transcript as it grows, measures the relationship between cached context and new prompt content, and reports whether carrying the current context is becoming costly. The project applies an EOQ-style model to prompt caching rather than attempting to compress or prune the conversation itself.
The dashboard presents context buckets for files, skills, and tools. Each path includes a token count, and selecting or clearing paths previews how the estimated restart cost changes. A terminal statusline can show a compact view of the current state.
The measurement and display layers are observers. Transcripts are read locally and are not modified or uploaded. Metrics are returned through the dashboard, statusline, or MCP tools instead of being inserted into the model's context.
The native input is a Claude Code JSONL session transcript. The pipeline extracts cache_read_input_tokens from turns, follows the transcript as new turns arrive, and derives measures such as context inventory, growth gap, EOQ-curve position, and bill premium. The dashboard uses green, amber, and red thresholds for the reported premium: below 10 percent, 10โ24 percent, and 25 percent or higher.
A session can be rotated when a restart is appropriate. Before clearing the old session, the handoff workflow can preserve selected paths, a summary, and the next task. A new session can load that package by its returned semantic token, by text search, or by matching the current project. Turn-history tools then inspect the lineage associated with the loaded handoff.
Node.js 22.16 or newer is required. For a quick read-only demonstration, run npx -y @nomadop/session-watcher demo. To inspect an existing transcript, use the replay command with the path to a Claude Code JSONL file.
The recommended installation is the Claude Code plugin:
claude plugin marketplace add nomadop/session-watcher.claude plugin install session-watcher@session-watcher./reload-plugins if the plugin was added or updated during an existing session.The plugin registers the MCP tools and a SessionStart hook that launches the dashboard server. Statusline support requires adding a command entry to Claude Code's ~/.claude/settings.json, pointing to the installed dist/statusline.js. The plugin cache path can change after updates, so the path may need to be found again.
The nomadop/session-watcher MCP server provides tools in four groups:
History tools resolve the lineage of the currently loaded handoff and do not take a separate lineage identifier. A cursor from turn_page confirms that more history exists; no cursor is not proof that the history is exhausted.
Native transcript support is for Claude Code. The project describes other agents as adapter-ready but pending, including OpenCode, OpenClaw, Hermes, and Aider. Adding another agent requires an extractor for its cache_read_input_tokens data.
The server measures the cost curve and supplies information for a restart decision; it does not replace pruning or compression tools. Only the explicitly selected handoff paths are rebuilt, and handoff is the mechanism that can inject preserved context into a later model session. No remote telemetry is used, and local transcript data remains local.
Factual signals from GitHub, npm, and our automated checks โ not a rating.
No reviews yet โ be the first to share how this listing worked for you.
Showcase your server listing on GitHub or your project documentation. Embed this dynamic SVG badge to highlight official listing status and live engagement.
[](https://allmcps.com/mcp/nomadop-session-watcher)<a href="https://allmcps.com/mcp/nomadop-session-watcher"><img src="https://allmcps.com/api/badge/nomadop-session-watcher?style=directory" alt="Session Watcher on AllMCPs" /></a>