- Adoption includes breadth, depth, quality, and trend—not only call count.
- Read tool usage inside completed and incomplete workflow paths.
- Treat descriptions and schemas as agent-facing UX that can be measured.
MCP tool adoption is not the same as call volume. A useful adoption measure includes how many distinct clients use a tool, whether usage is growing, whether calls succeed, and whether the tool participates in workflows that reach a useful outcome.
Measure adoption in three dimensions
- Breadth: how many distinct clients use the tool?
- Depth: how often is it used and how does it fit into a session path?
- Quality: do calls succeed and do the workflows that include the tool complete?
Start with the tool catalog
The catalog is the set of capabilities an agent can consider. Record which tools were advertised, when their descriptions or schemas changed, and which clients discovered them. A zero-call tool may be unnecessary, but it may also be invisible because its name or description does not match the user’s language.
The MCP tool adoption scorecard
- Reach: distinct clients calling the tool divided by active clients.
- Frequency: calls per active client, with outliers separated.
- Success: successful results divided by calls.
- Retry burden: repeated calls per workflow or per successful call.
- Workflow contribution: completed workflows containing the tool divided by started workflows containing it.
- Trend: week-over-week change, segmented by client and server version.
Read tool adoption inside workflows
A tool that starts many sessions but is followed by failure is not necessarily healthy adoption. Group calls into sessions, find common paths, and compare completed versus incomplete paths. The same tool can be valuable in one workflow and a dead end in another.
What to do with each pattern
- High reach, high success: protect the tool with regression tests and capacity planning.
- High reach, low success: fix the schema, validation, dependency, or error response first.
- Low reach, high success: improve naming, descriptions, examples, and placement in the catalog.
- High calls, low breadth: inspect automation or one noisy client before calling it broadly adopted.
- Zero calls: review discoverability and user need before deleting the tool.
- High adoption, low completion: inspect the next step in the path; the tool may be exposing a downstream failure.
How to measure a description change
Treat tool descriptions as an agent-facing UX surface. Record the old and new wording, the deployment date, affected clients, and the hypothesis. Compare breadth, success, retries, and workflow completion over a stable window. Avoid changing the schema, handler, model, and description at the same time if you want a useful result.
Define completion explicitly
The strongest completion event comes from the application: a ticket created, a deployment verified, or a report delivered. Emit a named workflow event with started, completed, and failed states. If that is not available, document the proxy used and keep it consistent over time.
Frequently asked questions
What is MCP tool adoption?
MCP tool adoption is the breadth, frequency, and quality of client usage for an MCP tool, evaluated in the context of sessions and workflow outcomes rather than raw call count alone.
How do I find unused MCP tools?
Count calls for each advertised tool over a meaningful period and compare the result with client coverage and catalog changes. Before removing a zero-call tool, check whether its description or schema prevents discovery.
What is MCP workflow completion rate?
Workflow completion rate is completed workflows divided by started workflows. Define completion using an actual business result when possible, or a documented proxy when the application outcome is not available.
See this on your own server
TrackMCP turns your MCP server's calls into adoption, workflows, and outcomes. One line to install.