feat(cli/mcp): add slash command prompts (/sessions, /spend) and updating docs - #4
Conversation
- Register 'sessions' and 'spend' MCP prompts so clients like Claude Code
surface /mcp__agentmeter__sessions and /mcp__agentmeter__spend as slash
commands; both accept optional string args (limit / days) and embed live
data directly into the prompt message for instant context
- Add prompts:{} to server capabilities
- README: add slash commands table, add Updating section covering npx @latest
vs global install and the restart-after-update note
|
Bugbot is not enabled for your account, so this pull request was not reviewed. Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs. |
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
Review
The README additions and MCP prompt structure look good — the slash command descriptions are accurate, the as const casts are correct, and the optional chaining on result.content[0]?.text properly handles noUncheckedIndexedAccess.
One issue to fix before merging: both prompt handler bodies await functions that touch the filesystem (handleListRecentSessions) or the network (handleGetMySpend) with no try/catch. Per CLAUDE.md, the CLI must never crash with an unhandled exception. Add a try/catch in each handler and return a graceful error message string — the MCP SDK can surface it to the client rather than the server crashing or returning a protocol-level error.
Generated by Agent: Code Review for issue #4 · ● 99.7K
| async ({ limit }) => { | ||
| const limitNum = limit ? Math.min(50, Math.max(1, Number.parseInt(limit, 10) || 10)) : 10; | ||
| const result = await handleListRecentSessions({ limit: limitNum }); | ||
| const data = result.content[0]?.text ?? '{}'; | ||
| return { | ||
| messages: [ | ||
| { | ||
| role: 'user' as const, | ||
| content: { | ||
| type: 'text' as const, | ||
| text: `Here are my ${limitNum} most recent AI coding sessions:\n\n${data}\n\nPlease summarise what I have been working on, call out any sessions that used an unusually high number of tokens, and give me the total token count across all sessions shown.`, | ||
| }, | ||
| }, | ||
| ], | ||
| }; | ||
| }, |
There was a problem hiding this comment.
Missing error handling — handleListRecentSessions involves file I/O and can throw. Per CLAUDE.md: "the CLI should never crash with an unhandled exception. Catch, log, continue."
Wrap the handler body in try/catch and return a graceful message on failure:
async ({ limit }) => {
try {
const limitNum = limit ? Math.min(50, Math.max(1, Number.parseInt(limit, 10) || 10)) : 10;
const result = await handleListRecentSessions({ limit: limitNum });
const data = result.content[0]?.text ?? 'No session data available.';
return {
messages: [{
role: 'user' as const,
content: { type: 'text' as const, text: `Here are my \$\{limitNum} most recent AI coding sessions:\n\n\$\{data}\n\n...` },
}],
};
} catch (err) {
return {
messages: [{
role: 'user' as const,
content: { type: 'text' as const, text: `Failed to fetch sessions: \$\{String(err)}` },
}],
};
}
},| async ({ days }) => { | ||
| const daysNum = days ? Math.min(365, Math.max(1, Number.parseInt(days, 10) || 7)) : 7; | ||
| const result = await handleGetMySpend({ days: daysNum }); | ||
| const data = result.content[0]?.text ?? '{}'; | ||
| return { | ||
| messages: [ | ||
| { | ||
| role: 'user' as const, | ||
| content: { | ||
| type: 'text' as const, | ||
| text: `Here is my AI coding spend data for the last ${daysNum} day${daysNum === 1 ? '' : 's'}:\n\n${data}\n\nPlease summarise my spending trends, highlight any notable spikes or patterns, and give me a sense of whether my usage is tracking high or low.`, | ||
| }, | ||
| }, | ||
| ], | ||
| }; | ||
| }, |
There was a problem hiding this comment.
Same missing error handling as the sessions handler — handleGetMySpend makes a network call and can throw. Same fix applies: wrap in try/catch and return a message on error rather than letting it propagate.
2 participants
Summary
Adds two MCP prompts so Claude Code and Cursor surface explicit slash commands, plus an Updating section in the README.
Slash commands
/mcp__agentmeter__sessionssessionslimit(optional, default 10)/mcp__agentmeter__spendspenddays(optional, default 7)Both prompts fetch live data via the existing handler functions and embed it directly in the user message — Claude has the data in context immediately without needing to decide to call a tool.
sessionsworks with no account.spendreturns the no-key hint gracefully if no API key is configured.README: Updating section
Covers three scenarios:
@agentmeter/cli@latestin config args — npx re-resolves on cache expirynpm install -g @agentmeter/cli@latestto update, useagentmeteras commandRelease
Merge and create a release with title
feat(cli/mcp): add slash command promptsto trigger a minor bump →1.1.0.