Tool catalog size
Which tools are exposed by default, which are task-specific, and which should be hidden, split, or lazy-loaded.
MCP token cost audit
MCP tools should help agents, not quietly fill the context window.
MCP can make coding agents dramatically more capable, but large tool catalogs and heavy schemas can add recurring token overhead to every session. Vorp Labs audits MCP servers and tool surfaces for teams trying to reduce cost without giving up capability.
Audit checks
Tool design is part of model cost engineering.
Schema weight
How much context is consumed by tool parameters, examples, enum values, descriptions, and response shapes.
Routing clarity
Whether agents can choose the right tool without reading overlapping names, vague descriptions, or redundant capabilities.
Result verbosity
Where tool responses return too much raw data instead of concise summaries, stable identifiers, or fetch-on-demand details.
Outcomes
Reduce overhead while keeping the useful tools.
- Lower baseline context cost before the agent starts real work.
- Cleaner tool selection and fewer failed calls.
- Better separation between always-on tools and specialized workflows.
- A roadmap for replacing tool calls with retrieval, deterministic code, or narrower commands where appropriate.
Part of the cost audit
MCP cleanup is usually one workstream, not the whole answer.
The broader audit also looks at project memory, prompt storage, model routing, and open-source model substitution so tool overhead is not optimized in isolation.
Request MCP review