Contact Us 1-800-596-4880
  1. Homepage
  2. Enhanced MuleSoft Experience

  3. Managing Costs and Token Usage

Managing Costs and Token Usage

Use cost and token metrics to understand usage, control spend, and identify optimization opportunities.

Available metrics depend on your tenant configuration and role.

Before You Begin

Before getting started, make sure you have:

  • An Anypoint Platform account.

  • These permissions:

    • API Manager: Manage Policies

    • API Manager: View APIs Configuration

    • API Manager: View Policies

    • Anypoint Code Builder: Mule Developer Generative AI User (Optional)

    • Anypoint Monitoring: Monitoring Viewer

    • Exchange: Exchange Viewer

    For more information, see Enhanced Experience Permissions.

Cost and Token Data Locations

The system displays token usage and performance data in several places:

  • Governance > Cost Management: Primary dashboard for organization-wide token usage and spend indicators

  • Service detail pages: Usage and performance summaries for supported services

  • Observability: Aggregated token and latency trends, when configured

Navigation labels can vary by catalog, enabled features, and release.

Open Cost Management

  1. Log in and go to Governance > Cost Management.

  2. Select a tab:

    • The Budgets tab (default): View total managed spend and budget breakdowns.

    • The Optimizations tab: Analyze MCP server token usage and optimization opportunities.

  3. Set the time range for the data you want to analyze. The time range selector (1H, 24H, 7D, or 30D; default 30D) sits next to the tabs and applies to both the Budgets and Optimizations tabs.

  4. Use the Tokens/Cost toggle to choose how metrics display. See View Metrics in Tokens or Cost.

View Metrics in Tokens or Cost

The Tokens/Cost toggle next to the tabs controls whether Cost Management metrics display as token counts or as dollar amounts. The toggle defaults to Tokens.

To view spend in dollars, set the toggle to Cost.

Before dollar amounts appear, you must configure a cost rate for each model. Until you set model rates, the Cost view shows a prompt to configure your models, and metrics that depend on cost show zero. To set rates, select Configure Costs, or set a cost per token for each model on the Models page. See Setting Model Costs for Spend Tracking.

When Cost Tracking Begins

Cost metrics accrue only from the time you configure model rates. LLM usage before that time isn’t priced and doesn’t appear as cost, even for earlier periods within the selected time range.

Token counts are unaffected. The Tokens view continues to reflect the complete spend from the time you enabled Cost Management, regardless of when you set model rates.

Cost Data Use Cases

Cost and performance data inform multiple workflows:

  • Cost optimization: Identify services with high usage and apply optimization controls.

  • Performance triage: Correlate latency or error spikes with deployments or policy changes.

  • Planning and alignment: Share trends with finance and platform teams for forecasting.

To automatically cap usage, configure budgets in the Budgets tab or through model wallets. Budgets block requests once the limit is reached. To limit consumption in other ways, apply governance strategies or rate-limiting policies.

Cost Data and Governance

Use Cost Management data to create or refine governance strategies.

For runtime enforcement, attach cost-related controls or policies to a governance strategy and apply it to the relevant services.

Budgets Tab

The Budgets tab shows financial summaries across your configured budgets:

  • Total Managed Spend: Total spend across all configured budgets.

  • By Provider: Spend breakdown by LLM provider.

  • By Model Wallet: Spend breakdown by model wallet.

These summaries display as token counts or dollar amounts based on the Tokens/Cost toggle. Dollar amounts appear only after you configure model rates. See View Metrics in Tokens or Cost.

Each budget row displays a status badge that reflects current usage against the budget limit:

  • At Risk: Usage has reached 75% or more of the budget limit.

  • At Limit: Usage has reached the budget limit. The system blocks further requests.

Use the search on the Budgets table to find specific budgets.

Add a Budget

  1. In the Budgets tab, select New Budget.

  2. In the New Budget dialog, configure the budget:

    • Model Wallet (required): Select the model wallet to scope the budget.

    • Provider (required): Select an LLM provider.

    • Model (required): Select a model. This field appears after you select a provider.

    • Period: Select Daily, Weekly, or Monthly.

    • Metric: Select Spend to set a dollar limit, or Tokens to set a token count limit.

    • Spend Limit (USD) or Token Limit (required): Enter the maximum allowed value.

  3. Select Create Budget.

When usage reaches the budget limit, the system blocks requests. Each model wallet, provider, and model combination supports only one budget.

For information about model wallets and how they combine authentication credentials with budget limits, see Capping Spend for Callers of Model Proxies.

Optimizations Tab

The Optimizations tab shows cost and token savings data across model proxies, agents, and MCP servers.

The summary cards show:

  • Total Managed Spend: The total cost of all LLM traffic routed through managed proxies. This excludes any usage that bypasses the proxy network.

  • Agents Potential Savings: Estimated cost savings available by applying optimization policies to agents. It shows Amount Saved as the amount already saved.

  • LLM Proxies Potential Savings: Estimated cost savings available by applying optimization policies to LLM proxies. It shows Amount Saved as the amount already saved.

  • MCP Servers Potential Savings: Estimated token savings available by applying optimization policies to MCP servers. It shows Tokens Saved as tokens already saved.

Treat these metrics as directional, not exact.

The Top optimization opportunities section highlights the instances with the highest potential savings and the specific optimization policy to apply.

Use the controls at the top of the Optimizations tab to narrow the view:

  • Select a sub-tab to filter by instance type:

    • Model Proxies: Monitor cost data and savings for traffic explicitly routed through your managed proxy network.

    • Agents: Track proxy-attributed spend and policy savings for active agents governed by Omni Gateway.

    • MCP Servers: Observe token volumes processed through tool responses during active server calls.

  • Search for instance names.

  • Use Filters to refine the view by Environment or Provider.

The counter updates to show the number of matching instances.

Apply Optimization Policies

Optimization policies reduce token consumption and cost for a given instance without changing what the instance does. The available policies depend on the instance type.

To apply an optimization policy:

  1. In the Optimizations tab, select the sub-tab for the instance type: Model Proxies, Agents, or MCP Servers.

  2. Locate the instance you want to optimize, and select View optimizations.

  3. In the instance’s optimizations view, locate the policy you want to apply.

  4. Review the policy’s Potential savings, and expand Why this policy? for details.

  5. Select Apply to enable the policy.

Treat the savings estimates as directional, and note that some policies may affect instance latency.

MCP Optimizations

For MCP servers, you can apply these optimization policies. Each policy shows its Potential savings and a Why this policy? explanation, and you apply it by selecting Apply.

  • MCP Tool Mapping: Redefines tool names and descriptions to reduce prompt complexity and improve LLM comprehension.

  • MCP Tools Progressive Disclosure: Wraps the MCP server and surfaces two tools, search_tools and invoke_tool, to increase both accuracy and cost optimization.

  • Clean Payloads: Strips structural noise from tool responses, such as whitespace, nulls, and base64 blobs.

  • Smart Response Trimming: Trims oversized responses to match agent intent, dropping fields the agent never reads.

  • Compress Repeated Structures: Compresses repeated tabular structures into Token-Oriented Object Notation (TOON).

Agent Optimizations

For agents, you can apply these optimization policies to the agent’s Model Proxy to reduce token costs. Each policy shows its Potential savings and a Why this policy? explanation, and you apply it by selecting Apply.

  • Clean Responses: Strips redundant fields and boilerplate from tool responses before they reach the model.

  • Compress Repeated Structures: Re-encodes repeated array or object structures in tool responses into a more compact form.

Model Proxy Optimizations

For model proxies, you can apply these optimization policies to reduce token costs. Each policy shows its Potential savings, and you apply it by selecting Apply. Applying policies may affect instance latency.

  • Semantic Caching: Caches LLM responses by semantic similarity so repeat or near-repeat prompts short-circuit an upstream call. This typically results in about 40% fewer tokens on chat-style traffic.

See Also