This MCP server provides AI cost tracking and budget management across Claude Code, Cline, Cursor, and Claude Desktop, supporting 11 LLM providers via two sets of tools.
Proxy Tools (API Key Required)
Usage stats: Query total spend, request counts, top models, and cache hit rates for today, week, or month
Cost breakdowns: Analyze costs grouped by provider, model, session, or day
Budget status: Check budget limits and remaining balances
Session summaries: Review recent proxy sessions with cost, duration, and models used
API key management: List all keys with status and creation dates
Health check: Ping the proxy and measure response time
Local Tools (No API Key Needed)
Current session cost: Aggregate real-time costs across all detected AI coding tools (Claude Code, Cline)
Project overview: View cumulative costs ranked by spend across all projects
Cache savings: Analyze how much prompt caching has saved you
Monthly forecast: Project monthly spending and compare against Claude Max subscription cost
Subagent attribution: Break down costs by subagent within a Claude Code session
Additionally, a SessionEnd hook automatically logs session costs when Claude Code exits. Data is auto-detected and aggregated from Claude Code, Cline (VS Code, Cursor, Windsurf), WSL installations, and remote VS Code/Cursor Server extensions.
Automatically detects and aggregates cost data from AI coding tools used within VSCodium, offering insights into session costs, project spending, and prompt caching savings.
AI cost inspection for supported Claude Code sessions and Cline task data found in VS Code-family storage. Eleven tools cover local session evidence, authenticated gateway spend, and budget queries.
Part of LLMKit, an open-source API gateway with cost tracking and budget enforcement.
Add to your .mcp.json (Claude Code) or .cursor/mcp.json (Cursor):
{
"mcpServers": {
"llmkit": {
"command": "npx",
"args": ["-y", "@f3d1/llmkit-mcp-server"]
}
}
}The local tools (llmkit_local_*) need no API key. They read supported Claude Code sessions and
Cline task data from supported editor storage. Proxy tools require an existing LLMKit API key in
LLMKIT_API_KEY. Check llmkit.sh for current account and service availability.
Related MCP server: Langfuse MCP Server
Tool | What it does |
| Spend, requests, top models for a period |
| Costs grouped by provider, model, session, or day |
| Budget limits and remaining balance |
| Recent sessions with cost, duration, models |
| All keys with status and creation date |
| Proxy ping with response time |
The local tools detect supported installations and aggregate their session data.
Tool | What it does |
| Current Claude Code session or latest detected Cline task cost |
| Cumulative cost across all projects and sessions |
| Claude Code prompt caching savings using model-bound prices |
| 30-day API-rate projection from detected local history |
| Subagent cost attribution (Claude Code) |
Auto-log session costs when Claude Code exits:
{
"hooks": {
"SessionEnd": [
{
"type": "command",
"command": "npx @f3d1/llmkit-mcp-server --hook"
}
]
}
}Variable | Required | Description |
| No | API key for proxy tools. Local tools work without it. |
| No | Proxy URL (defaults to hosted service) |
| No | Override Cline data directory path |
| No | Set to |
The local tools read data from:
Claude Code (~/.claude/projects/)
Cline extension storage in VS Code, Insiders, VSCodium, Cursor, and Windsurf
WSL installations when LLMKIT_SCAN_WSL=1 (scans distro homes via UNC paths on Windows)
VS Code, Cursor, and Windsurf server directories for supported remote extension storage
MIT
Check budget limits and remaining balance
Inputs: budgetId. Returns: budgets
Query cost breakdown grouped by provider, model, session, or day
Inputs: groupBy (required), days, provider, model. Returns: groupBy, days, breakdown
Check proxy health and response time
Inputs: no named parameters. Returns: status, responseTimeMs
List all API keys with status and creation date
Inputs: no named parameters. Returns: keys
Subagent cost attribution for the current Claude Code session. Shows which agents cost the most.
Inputs: no named parameters. Returns: sessionId, totalCostUsd, mainConversationCostUsd, subagentsTotalCostUsd, +2 more
Cache savings analysis across all detected AI coding tools. Shows how much prompt caching saved.
Inputs: no named parameters. Returns: savings, totalSavedUsd
Monthly cost projection based on local AI tool usage. Compares to Max subscription.
Inputs: no named parameters. Returns: projectedMonthlyUsd, dailyAverageUsd, totalTrackedUsd, totalSessions, +1 more
Cumulative cost across all projects and sessions from all detected AI coding tools, ranked by spend.
Inputs: no named parameters. Returns: projects, totalCostUsd
Current session cost across all detected AI coding tools (Claude Code, Cline). No API key needed.
Inputs: no named parameters. Returns: sessions, totalCostUsd, sourceCount
Get recent proxy sessions with cost, duration, and models used
Inputs: sessionId, limit. Returns: sessions
Get usage statistics (spend, requests, top models) for a time period
Inputs: period. Returns: period, requests, totalSpendUsd, inputTokens, +4 more
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 11 tools
Tools are mostly distinct, targeting different aspects like budget, cost breakdown, health, keys, local session, etc. However, llmkit_cost_query, llmkit_usage_stats, and llmkit_session_summary have overlapping purposes (cost/spend/usage data) which could cause some confusion.
All tools follow a consistent pattern with the 'llmkit_' prefix and descriptive noun phrases separated by underscores, e.g., llmkit_budget_status, llmkit_cost_query. No mixing of conventions.
With 11 tools, the count is well within the recommended 3-15 range. Each tool covers a distinct aspect of proxy cost monitoring and management, justifying its inclusion.
The tool surface covers major monitoring needs: health, budget, cost, keys, sessions, and local attribution. However, it lacks management capabilities (e.g., create/update keys or budgets) and could include actions like setting usage limits, which are minor gaps for a read-only analytics server.
Query cloud, AI and SaaS spend across 25+ providers: costs, budgets, anomalies, unit economics.
Track API spend, credit balances and runway across OpenAI, Anthropic, OpenRouter and 25 more.
FinOps assistant: query AWS, GCP and Azure cost, commitments, budgets and reports.
Ask about AWS, Azure, GCP, OpenAI, Anthropic and Snowflake costs, budgets, and anomalies.
