MCP Setup — LLM Pricing API

Protocol

Endpoint

POST https://llm-pricing-api.fly.dev/mcp

Protocol

MCP 2024-11-05 · JSON-RPC 2.0

Tools

29 tools · No auth required

Client Setup

Recommended. No installation required — your client connects directly to the hosted server at https://llm-pricing-api.fly.dev/mcp.

Add to your client's MCP config

Use the url field — no command or local path needed.

{
  "mcpServers": {
    "llm-pricing": {
      "url": "https://llm-pricing-api.fly.dev/mcp"
    }
  }
}

Restart your client

After saving the config, restart Claude Desktop / VS Code / Cursor. The 29 tools will appear automatically.

Ask a question

Try: "What is the cheapest model for RAG under $2/1M tokens?"

Option A (remote, easiest): use the Remote HTTP tab above. Option B below runs a local Python server — works offline.

Clone and install

git clone https://github.com/skakumanu/llm-pricing-mcp-server.git
cd llm-pricing-mcp-server
python -m venv .venv
# macOS/Linux:  source .venv/bin/activate
# Windows:        .venv\Scripts\Activate.ps1
pip install -r requirements.txt

Edit `claude_desktop_config.json`

Windows: %APPDATA%\Claude\claude_desktop_config.json macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "llm-pricing": {
      "command": "python",
      "args": ["-u", "/absolute/path/to/llm-pricing-mcp-server/mcp/server.py"],
      "cwd": "/absolute/path/to/llm-pricing-mcp-server",
      "env": { "PYTHONUNBUFFERED": "1" }
    }
  }
}

Restart Claude Desktop

Tools appear in Claude's tool list. You'll see a hammer icon when the MCP server is active.

VS Code has two separate MCP setups depending on your extension — pick the one you use. Copilot reads its own servers config; Cline / Continue use a mcpServers config in their own settings UI.

One-click install

Add to VS Code

Opens VS Code directly. If nothing happens, your VS Code version doesn't support MCP deep links yet — use the JSON below instead.

Open your MCP config

Command Palette (Ctrl+Shift+P) → MCP: Open User Configuration for all workspaces, or create .vscode/mcp.json in this workspace for a project-only server.

Add the server

Copilot's native MCP config uses a servers key with an explicit type, not the mcpServers shape other clients use:

{
  "servers": {
    "llm-pricing": {
      "type": "http",
      "url": "https://llm-pricing-api.fly.dev/mcp"
    }
  }
}

Enable the tools in Agent mode

Open the Copilot Chat panel, switch to Agent mode, then click the tools icon and enable llm-pricing. All 29 tools become available to the agent.

Open your MCP config

In Cline: Settings (⚙) → MCP Servers → Edit JSON. In Continue: open your MCP server config (mcpServers section of Continue's settings).

Add the server

{
  "mcpServers": {
    "llm-pricing": {
      "url": "https://llm-pricing-api.fly.dev/mcp"
    }
  }
}

Reload the extension

Open the command palette (Ctrl+Shift+P) → Cline: Reload MCP Servers (or restart VS Code for Continue). Tools will appear in the agent panel.

One-click install

Add to Cursor

Opens Cursor directly. If nothing happens, your Cursor version may not support MCP deep links yet — use the JSON below instead.

Open Cursor MCP config

Windows: %USERPROFILE%\.cursor\mcp.json macOS/Linux: ~/.cursor/mcp.json

Add the server

{
  "mcpServers": {
    "llm-pricing": {
      "url": "https://llm-pricing-api.fly.dev/mcp"
    }
  }
}

Restart Cursor

Go to Settings → Features → MCP to confirm the server shows as connected.

Open Perplexity MCP config

Windows: %APPDATA%\Perplexity\mcp_config.json macOS: ~/Library/Application Support/Perplexity/mcp_config.json Tip: verify the exact path in Perplexity → Settings → MCP Servers.

{
  "mcpServers": {
    "llm-pricing": {
      "url": "https://llm-pricing-api.fly.dev/mcp"
    }
  }
}

Restart Perplexity

After saving, restart the desktop app. All 29 pricing tools will be available in your conversations.

Verify

Ask: "What LLM pricing tools do you have available?" — Perplexity should list all 29 tools.

Available Tools

All 29 tools are accessible via MCP. Purple = required arg · Grey = optional.

get_all_pricing

Live pricing for all 150+ models across 26 providers.

No arguments required

estimate_cost

Cost breakdown for a single model and token count.

model_nameinput_tokensoutput_tokens

compare_costs

Side-by-side cost comparison across multiple models.

model_names[]input_tokensoutput_tokens

predict_cost

Paste a prompt — get every model ranked cheapest-first. Counts tokens, infers the task, estimates output length, and factors in prompt-cache savings.

prompttask_typecache_hit_ratiotop_nrequire_function_callingrequire_visionmin_context_tokens

optimize_workload

Route each task to its own cheapest qualifying model instead of one model for everything. Reports the saving versus a single-model deployment.

workloads[]monthly_budget_usdmin_quality_score

recommend_model

Describe your use case in plain English — get the optimal model recommendation plus up to 3 alternatives. Classifies the description into a task type, then runs the same routing engine behind /router/recommend.

descriptiontask_typemax_cost_per_1m_tokensmin_quality_scoremin_context_windowpreferred_providermonthly_budget_usd

check_price_drift

Audit our own pricing against an external reference registry. A drifted price is automatically withheld from serving the moment it's detected — this reports which models and why. The price value itself is never overwritten.

threshold_pctproviderlimit

get_data_quality

Aggregate accuracy summary for the whole catalogue: % confirmed and fresh, how many models are withheld for price drift, how many are unpriced or stale. A single trust signal, distinct from check_price_drift's per-model detail.

No arguments required

get_ide_pricing

Subscription pricing for AI coding IDEs: Copilot, Cursor, Windsurf, JetBrains AI, Amazon Q, Tabnine.

providermax_monthlyinline_only

get_performance_metrics

Throughput, latency, context window, and quality scores.

providerinclude_cost

get_use_cases

Model recommendations by use case and task type.

provider

get_telemetry

Server usage analytics and tool call statistics.

include_detailslimit

get_pricing_history

Historical pricing snapshots for any model or provider.

model_nameproviderdayslimit

get_pricing_trends

Price-change leaderboard — biggest movers over N days.

dayslimit

register_price_alert

Webhook alert when a model price changes by threshold%.

urlthreshold_pctprovidermodel_name

list_price_alerts

List all registered price-change webhooks.

No arguments required

delete_price_alert

Remove a registered webhook by ID.

alert_id

get_pricing_export_url

Generate a CSV or JSON export URL for pricing data.

formatmodel_nameproviderdays

list_conversations

List recent agent conversation sessions.

limit

delete_conversation

Delete a conversation by session ID.

conversation_id

ask_agent

Ask the ReAct AI pricing agent a natural-language question.

messagesession_id

record_usage

Record actual token usage for a completed LLM call. Cost is computed server-side from current pricing, never trusted from the caller — the real-spend counterpart to estimate_cost. Optionally tag with session_id for later per-session analysis.

model_nameinput_tokensoutput_tokensorg_idoccurred_atrequest_idsession_id

get_usage_summary

Summarize actual recorded spend: total cost, request count, and breakdowns by model and provider, from usage previously logged via record_usage.

org_iddays

analyze_session_usage

Analyze one session's actual recorded usage on demand: total tokens, total cost, per-model breakdown, and a grounded model recommendation with estimated savings — or confirmation the current choice is already optimal.

session_id

register_budget_alert

Register a webhook to be notified when actual recorded spend crosses a USD threshold within a rolling window. Fires at most once per period_days while spend stays over threshold.

urlthreshold_usdorg_idperiod_days

list_budget_alerts

List all registered budget-spend webhook alerts with their IDs and settings.

No arguments required

delete_budget_alert

Delete a registered budget-spend alert by its ID.

alert_id

Sample Prompts

What is the cheapest model for RAG pipelines under $2/1M tokens?

Compare GPT-4o vs Claude Sonnet vs Gemini 1.5 Pro on cost and quality

What would 10M tokens per month cost on each major provider?

Which models have dropped in price in the last 30 days?

Which model gives the best quality per dollar for coding tasks?

Click any prompt to copy it.

Troubleshooting

Problem	Fix
Tools not showing after restart	Check the config file path and JSON syntax. Use a JSON validator if unsure. Ensure no trailing commas.
STDIO server hangs with no output	Expected — the STDIO server waits for JSON-RPC input on stdin. It is not a web server; use it only with a MCP client.
`ModuleNotFoundError` (local STDIO)	Activate the virtualenv (`source .venv/bin/activate`) and run from the repo root directory.
`Model not found` error	Call `get_all_pricing` first to get valid model names — they are case-sensitive.
Connection refused / timeout	Run `curl https://llm-pricing-api.fly.dev/health` to verify the server is up. If using local STDIO, ensure the server process is running.
`401 Unauthorized` on MCP endpoint	The `/mcp` endpoint is public — no API key needed. If you see 401, check for a misconfigured auth header in your client.

MCP Setup Guide

Add to your client's MCP config

Restart your client

Ask a question

Clone and install

Edit claude_desktop_config.json

Restart Claude Desktop

One-click install

Open your MCP config

Add the server

Enable the tools in Agent mode

Open your MCP config

Add the server

Reload the extension

One-click install

Open Cursor MCP config

Add the server

Restart Cursor

Open Perplexity MCP config

Restart Perplexity

Verify

Edit `claude_desktop_config.json`