MCP Pricing

MCP pricing estimates the cost of tool traffic. Unlike model pricing, MCP pricing is based on calls and bytes rather than tokens.

The UI accepts human-readable values:

USD per 1M calls
USD per 1M input bytes
USD per 1M output bytes

Odock stores the values internally as nanos USD per call or byte, then uses them to estimate each MCP request.

Pricing Fields

UI field	Internal field	Meaning
Price (USD / 1M calls)	`callNanosUsd`	Flat cost for each MCP request.
Input price (USD / 1M bytes)	`inputByteNanosUsd`	Cost for bytes sent from the caller to the MCP server.
Output price (USD / 1M bytes)	`outputByteNanosUsd`	Cost for bytes returned by the MCP server.

Cost Formula

estimated cost =
  flat call cost
  + input byte cost
  + output byte cost

Example

Assume an MCP server has:

Field	Value
Price per 1M calls	`$2.00`
Input price per 1M bytes	`$0.50`
Output price per 1M bytes	`$1.00`

For one call with 2,000 input bytes and 8,000 output bytes:

call cost   = 2.00 / 1,000,000 = $0.000002
input cost  = 0.50 * 2,000 / 1,000,000 = $0.000001
output cost = 1.00 * 8,000 / 1,000,000 = $0.000008
total       = $0.000011

What Usage Records Store

MCP usage records include:

MCP server id
transport
method
tool name
input bytes
output bytes
pricing snapshot
total cost in nanos USD
status and latency

The pricing snapshot matters because future pricing changes should not rewrite historical usage cost.

Pricing Strategy

Use different MCP pricing when:

A tool calls a paid external API.
A tool performs expensive internal work.
A client contract needs tool-specific chargeback.
A team needs cost visibility even when tools do not use model tokens.

For UI steps, see Edit MCP pricing.

To review the pricing for an MCP call see MCP Usage Records.