APIpie AI

@APIpie list all available vision models from Anthropic and OpenAI, showing their context windows and pricing

Found 6 vision-capable models matching your criteria. Anthropic Claude 3.5 Sonnet supports 200K context at $3/$15 per million tokens (input/output). Claude 3 Opus offers 200K context at $15/$75. OpenAI GPT-4o provides 128K context at $2.50/$10, while GPT-4o mini offers 128K at $0.15/$0.60. GPT-4 Turbo supports 128K context at $10/$30. All models accept image inputs alongside text. This is an illustrative summary—actual availability and pricing may vary by region and account tier.

This example demonstrates the List AI Models tool's filtering capabilities. Useful for comparing options before committing to a provider. Note that model availability changes frequently, and some models may have country restrictions you'll need to verify separately using the fetch restrictions tool.

@APIpie show me the last 20 API calls from my account, breaking down total cost and average latency by model

Retrieved 20 queries from the past 48 hours. Total cost: $2.47. GPT-4o accounted for 12 calls ($1.89, avg 847ms latency). Claude 3.5 Sonnet had 6 calls ($0.52, avg 1,203ms). Gemini 1.5 Pro had 2 calls ($0.06, avg 612ms). Highest single query cost was $0.34 (GPT-4o, 18.2K output tokens). All requests originated from IP 203.0.113.45. This represents a sample query history view—your actual logs will reflect your usage patterns and may include additional metadata.

The Get query history tool enables cost tracking and performance auditing across providers. Requires valid API key authentication. Particularly valuable for teams managing multi-model workflows or optimizing spend. Latency data helps identify which models best fit your response-time requirements.

@APIpie delete all vectors in the 'customer-feedback-q1' collection since we've migrated to a new schema

Confirmed deletion request for collection 'customer-feedback-q1'. This action will permanently remove all vector embeddings and associated metadata from this collection. The operation is irreversible. If you intended to delete specific vectors rather than the entire collection, you can target individual vector IDs instead. Proceed with caution—this example illustrates the deletion workflow, but in production you should verify collection names and backup critical data before executing.

The Delete Vectors tool handles both full collection removal and selective vector deletion. This is a destructive operation with no undo mechanism. Best practice: confirm collection names carefully and maintain backups of vector data if recovery might be needed. Useful when rotating embeddings or cleaning up deprecated datasets.

When APIpie wins for client billing transparency

A 6-person agency runs client projects across GPT-4, Claude, and Gemini, billing model usage back to each account. APIpie's query history tool surfaces token counts, costs, and source IP per request, so the ops lead can reconcile invoices without spreadsheet archaeology. The model restrictions tool confirms which clients can deploy in EU-only regions before quoting. This works if you're routing through APIpie's gateway already—if you're calling OpenAI direct, you're adding a proxy layer just for logs, which only pays off above $2k/month in model spend. For teams under that threshold, a simple usage dashboard in your existing provider is faster. If you're multi-provider and need per-request attribution, APIpie closes the gap between raw API logs and actual client billing.

When this MCP speeds up voice feature prototyping

A 3-person startup is building a voice assistant demo for Series A pitch meetings and needs to test ElevenLabs, Play.ht, and Azure voices in one afternoon. The List AI Models tool filters by voice subtype and returns the current catalog without hunting through three vendor docs. The fetch restrictions tool confirms which voices work in the founder's target markets before the demo script is finalized. This is a win if you're genuinely comparing providers—if you've already picked ElevenLabs, you don't need a meta-catalog. The vector deletion tool is irrelevant here unless you're also embedding user transcripts, which most early demos skip. For teams that prototype across providers weekly, APIpie cuts model research from an hour to five minutes. If you build on one stack, it's overkill.

When query logs matter for SOC 2 evidence

A 12-person SaaS company is prepping for SOC 2 and needs to prove that AI model calls respect data residency rules and log access patterns. The get query history tool exports latency, token counts, and source IPs for the auditor's sampling period. The model restrictions tool documents which models were available in compliant regions at the time of each call. This works if APIpie is your gateway and you're already logging through it—retrofitting this MCP onto direct provider calls won't generate the history. The vector tools are a distraction unless your product embeds user data, which adds a separate compliance surface. If you're under 10k API calls per month, your existing provider's logs are probably sufficient. Above that, APIpie's structured history saves the engineering team from writing custom log parsers for audit season.