cargo-mcp defines 12 tools with consistent naming (verb_noun pattern: cargo_check, cargo_build, etc.) and reasonable descriptions. All tools have input schemas with type definitions and parameter descriptions. However, several critical gaps reduce quality: (1) Output schemas are NOT documented, tools return execution results but the response structure is undefined for LLMs; (2) Error handling lacks actionable recovery guidance, tools will fail (e.g., 'Cargo.toml not found') but responses don't guide the LLM's next step; (3) Parameter constraints are minimal, many accept free-form strings (e.g., 'features', 'args') without format specifications; (4) Descriptions are functional but brief (avg ~100 chars) and lack WHEN/WHY context; (5) Tool descriptions do not explicitly state which tools are idempotent vs. destructive (though risk classification is provided separately). The server is above-average community quality but lacks polish for production recommendation.
Add dependencies to Cargo.toml using cargo add
Run cargo bench to execute benchmarks
Build the project with cargo build
Run cargo check to verify the code compiles
Remove artifacts that cargo has generated in the past
Run cargo clippy for linting suggestions
Check if code is properly formatted without modifying files
Output schemas are completely undocumented. Tools return execution results (stdout, stderr, exit codes), but the response structure is not formally defined. LLMs cannot plan downstream tool calls or reliably extract success/failure signals.
Error handling lacks actionable recovery guidance. Tools will fail (e.g., 'Cargo.toml not found', network timeouts), but error responses do not tell the LLM what to do next (retry, ask user, call a different tool).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 63 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 58 | - | v1 |
Remove dependencies from Cargo.toml using cargo remove
Run a binary or example from the current package
Run cargo test to execute tests
Update dependencies using cargo update
Set the working directory for cargo operations. This sets the shared working directory that will be used by all AI tools, not just cargo-mcp. Other MCP servers like fs-mcp will also use this directory.
Parameter 'features' (in cargo_run) and 'args' (in cargo_run) accept free-form strings without format specification. The description says 'space-separated list' but does not enforce or validate the format. LLMs may pass malformed input.
Parameter 'cargo_env' (object type) is present on most tools but has a vague description: 'Optional environment variables to set for the cargo command'. No schema for the object structure; LLMs may pass invalid key-value formats. Should specify: 'Object with string keys and string values, e.g. {"RUST_BACKTRACE": "1"}'.
Tool descriptions do NOT mention idempotency, retryability, or destructiveness inline. While risk metadata is provided separately (READ_ONLY, WRITE, DESTRUCTIVE), LLM-facing descriptions should say 'This tool is idempotent: safe to retry' or 'This tool modifies Cargo.toml; confirm before running in production.' Baseline A+ tools embed this context.
set_working_directory tool description states it sets 'the shared working directory' but does not explicitly explain: (1) Why LLMs must call this first (all other tools require it); (2) Whether the path must be a Rust project or any directory; (3) How it interacts with cross-server communication. Description is 221 chars but lacks WHEN/WHY context.