EN
Documentation / MCP server

MCP server

Connect any MCP-capable agent to live LLM prices: six read-only tools over Streamable HTTP. No key required on the free tier.

Endpoint & keys

Server URL: https://techagent.net/mcp · transport: Streamable HTTP (stateless, JSON responses)

  • Endpoint: POST https://techagent.net/mcp - Streamable HTTP transport, stateless, plain JSON responses.
  • All 6 tools are read-only (readOnlyHint: true) and idempotent; the server only reads its own database.
  • No key is required on the free tier: anonymous usage is limited to 60 requests/hour per IP.
  • A pr_live_ key raises the limits; pass it as Authorization: Bearer pr_live_… or the ?key= query fallback. The quota is shared with the REST API.
  • Every tool accepts response_format: 'markdown' (default, compact text) or 'json'; structuredContent always carries the full machine-readable envelope.
  • Data errors come back as tool execution errors (isError: true) with a machine-readable error object {code, message, next_step} - including RATE_LIMITED with retry_after_seconds.

Connect

Claude Code (CLI):

claude mcp add --transport http priceradar https://techagent.net/mcp

Any client with an mcpServers config (e.g. .mcp.json):

{
  "mcpServers": {
    "priceradar": {
      "type": "http",
      "url": "https://techagent.net/mcp"
    }
  }
}

claude.ai (custom connector):

  1. Open claude.ai → Settings → Connectors → Add custom connector.
  2. Paste the server URL: https://techagent.net/mcp
  3. Save - no API key is needed on the free tier.

Tools

priceradar_list_models

Search and filter AI models with current API prices in USD per 1M tokens. Supports free-text query, provider/modality/status/channel filters, min_context and max_input_price bounds, five sort keys and keyset pagination (limit ≤ 100). The entry point for discovering valid model ids.

priceradar_get_model

Fetch one model's card by its stable id provider:model-slug: current prices across all channels, context window, lifecycle status and optional sections (pricing, deprecation, benchmarks, aliases, history_summary). Unknown ids fail with did-you-mean suggestions instead of an empty result.

priceradar_price_history

Published price-change events for one model over a date window of at most 365 days per call. granularity='changes' returns only the actual change events; 'daily' expands them into a daily price series suitable for charting.

priceradar_compare

Side-by-side comparison of 2-5 models: input/output/blended prices, context window, benchmark score and price per benchmark point. Fails with suggestions if any id is unknown, so agents can self-correct in one step.

priceradar_estimate_cost

Estimate the monthly USD cost of a token workload. Pass a task preset (chatbot, summarization, extraction, coding_agent, translation, rag_qa, classification) or explicit token volumes per request, plus optional cache_hit_rate and batch_share. Returns a per-model breakdown for the given model_ids, or the 10 cheapest active text models if omitted; all assumptions are echoed back.

priceradar_recent_changes

Feed of published pricing and lifecycle events - price cuts and rises, deprecations, retirements, launches, limit changes - newest first, filterable by window, event type, provider and minimum significance. Same data as the public RSS/JSON change feed.

REST mirror of the same tools: REST API reference

MCP server · PriceRadar