Tokencost Dev

io.github.jeff-atriumnPublished by GitHub account jeff-atriumnv0.1.3Updated Oct 24, 2026

LLM pricing oracle — model lookup, cost estimation, and comparison via LiteLLM

VerifiedSTDIODesktop onlyFinanceAI & ML

Overview

AI-generated overview

Local MCP server that gives an assistant LLM pricing data: model lookup, token cost estimates, and cost-based model comparison from the LiteLLM registry.

What you can do
  • Look up a model's per-token pricing, context window and capabilities, with fuzzy name matching
  • Estimate the cost of a given number of input and output tokens for a model
  • Compare models by cost-effectiveness, filtered by provider and mode
  • Force a refresh of the locally cached pricing data
When to use it
Useful when you want quick, concrete answers about LLM API prices, budget estimates for a prompt or workload, or help choosing the cheapest model that fits your requirements, without leaving the assistant.
Requirements
Runs as a local process via npx (Node.js required). No API keys, accounts or configuration files; pricing data is fetched from the LiteLLM community registry and cached locally for 24 hours, so network access is needed for fetching and refreshing.
Before you install
Prices come from the LiteLLM community registry and may lag behind official provider pricing, so verify before relying on estimates for billing decisions. The server fetches data from a third-party registry over the network.

Installation

In SourceWeft

  1. Open Tokencost Dev in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

[Image]

tokencost

[npm version] [npm downloads] [CI status] [license]

Ask your AI assistant "how much does GPT-4o cost?" — get an instant, accurate answer.

[tokencost demo in Claude Code]

Install in 30 seconds

Claude Code:

bash
claude mcp add tokencost-dev -- npx -y tokencost-dev

Then ask: "How much would 1M input tokens cost on claude-sonnet-4-5?"

Cursor (.cursor/mcp.json):

json
{  "mcpServers": {    "tokencost-dev": {      "command": "npx",      "args": ["-y", "tokencost-dev"]    }  }}

No API keys. No accounts. No configuration files. Pricing data is fetched from the LiteLLM community registry and cached locally for 24 hours.

Tools

get_model_details

Look up pricing, context window, and capabilities for any model. Fuzzy matching means "sonnet 4.5" works just as well as "claude-sonnet-4-5-20250514".

> "What are Claude Sonnet 4.5's pricing and capabilities?"
Model: claude-sonnet-4-5Provider: anthropic | Mode: chat
Pricing (per 1M tokens):  Input:  $3.00  Output: $15.00
Context Window:  Max Input:  200K  Max Output: 8K
Capabilities: vision, function_calling, parallel_function_calling

calculate_estimate

Estimate cost for a given number of input and output tokens.

> "How much will 1000 input + 500 output tokens cost on Claude Sonnet 4.5?"
Cost Estimate for claude-sonnet-4-5
  Input:  1K tokens × $3.00/1M  = $0.003000  Output: 500 tokens × $15.00/1M = $0.007500  ─────────────────────────────  Total:  $0.0105

compare_models

Find the most cost-effective models matching your requirements.

> "What are the cheapest OpenAI chat models?"
Top 2 most cost-effective models (provider: openai) (mode: chat):
1. gpt-4o-mini   Provider: openai | Mode: chat   Input: $0.15/1M | Output: $0.60/1M   Context: 128K in / 16K out
2. gpt-4o   Provider: openai | Mode: chat   Input: $5.00/1M | Output: $15.00/1M   Context: 128K in / 16K out

refresh_prices

Force re-fetch pricing data from the LiteLLM registry (cache is refreshed automatically every 24h).

Docs

Full documentation at tokencost.dev

License

MIT

Source: README.md at commit adb73fc

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v0.1.3LatestSep 16, 2026

Score

50% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts