fix(mcp): review fixes, multi-platform docs, remove Smithery

[GiggleLiu]

Summary

Review and cleanup of the MCP server implementation (commits 2b2e112..3ddc415 pushed directly to main).

Code Fixes

• Fix Box::leak memory leak in find_path_inner — replaced with CustomCost closure
• Remove redundant double-serialization (to_value then to_string_pretty) at 3 call sites
• Extract shared helpers in tools.rs — ser_vertex_weight_problem, ser_edge_weight_problem, ser_kcoloring
• Refactor integration tests — extract shared spawn_mcp/send/read_response/initialize/shutdown helpers; fix stdin ownership deadlock

Documentation

• Multi-platform MCP configs — add Cursor, Windsurf, OpenCode setup instructions to CLI help, README, and mdBook docs
• Remove Smithery artifacts — smithery.yaml, Dockerfile, and all Smithery references (Smithery requires a running server URL, incompatible with our stdio-based server)

Files Changed

• problemreductions-cli/src/mcp/tools.rs — memory leak fix, DRY helpers, serialization cleanup
• problemreductions-cli/tests/mcp_integration.rs — refactored test helpers, deadlock fix
• problemreductions-cli/src/cli.rs — updated MCP help text
• README.md — updated MCP section
• docs/src/mcp.md — removed Smithery, kept multi-platform configs
• .claude/CLAUDE.md — added make mcp-test
• Deleted: smithery.yaml, Dockerfile

Test plan

• make mcp-test passes (unit + integration)
• make clippy passes
• cargo build -p problemreductions-cli --release succeeds
• Manual MCP testing per issue Manual end-to-end testing: MCP server (pred mcp) #94

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.36%. Comparing base (3ddc415) to head (8e971cf).
⚠️ Report is 3 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #93   +/-   ##
=======================================
  Coverage   96.36%   96.36%           
=======================================
  Files         197      197           
  Lines       27210    27214    +4     
=======================================
+ Hits        26220    26224    +4     
  Misses        990      990           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

This PR refines the MCP server implementation in problemreductions-cli to address prior review findings, focusing on correctness for long-running operation and reducing duplicated serialization/test logic.

Changes:

• Replaced Box::leak-based cost selection for find_path with a CustomCost closure to avoid per-request memory leaks.
• Removed redundant JSON double-serialization patterns and extracted shared problem-serialization helpers for MCP instance creation.
• Refactored MCP integration tests with reusable helpers and adjusted subprocess I/O ownership to avoid deadlocks; ran rustfmt across touched files.

Reviewed changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
src/rules/graph.rs	Pure formatting adjustment in neighbor tree construction.
problemreductions-cli/tests/mcp_integration.rs	Extracted MCP subprocess/test helpers; improved shutdown behavior by taking ownership of ChildStdin.
problemreductions-cli/tests/cli_tests.rs	Formatting-only changes to CLI tests.
problemreductions-cli/src/mcp/tools.rs	Core MCP server fixes/refactors: CustomCost for path costs, DRY serialization helpers, removed double-serialization.
problemreductions-cli/src/mcp/prompts.rs	Formatting-only import consolidation.
problemreductions-cli/src/main.rs	Formatting-only change in command dispatch.
problemreductions-cli/src/commands/solve.rs	Minor formatting/closure style change.
problemreductions-cli/src/commands/graph.rs	Formatting-only change to color_fns initialization.
problemreductions-cli/src/commands/create.rs	Formatting-only change to argument extraction.
examples/reduction_satisfiability_to_circuitsat.rs	Formatting-only example output tweaks.
examples/reduction_maximumsetpacking_to_maximumindependentset.rs	Formatting-only example output tweaks.
examples/reduction_ksatisfiability_to_satisfiability.rs	Formatting-only example output tweaks.
.claude/skills/review-implementation/SKILL.md	Documentation update: adds DRY/KISS review checklist items.

Comments suppressed due to low confidence (2)

problemreductions-cli/src/mcp/tools.rs:350

• When cost is minimize:<field>, the code still chooses best_path by comparing p.len() (step count). That ignores the requested cost function across different source/target variant combinations, so the returned path may not be globally cheapest for the selected field. Consider tracking and comparing the total path cost (e.g., sum cost_fn.edge_cost over graph.path_overheads(p) while updating ProblemSize through each overhead), or changing find_cheapest_path to also return the final cost so the caller can pick the true minimum.

                if let Some(p) = found {
                    let is_better = best_path.as_ref().is_none_or(|bp| p.len() < bp.len());
                    if is_better {
                        best_path = Some(p);
                    }

problemreductions-cli/tests/mcp_integration.rs:45

• read_response loops until it can parse a JSON-RPC line, with no timeout or max-iteration guard. If pred mcp hangs or stops producing valid JSON (e.g., crashes after spawning, or writes unexpected output), this test will block indefinitely in CI. Consider adding a time limit (e.g., read in a separate thread and recv_timeout, or cap the number of lines / elapsed time before failing with a helpful panic message).

    fn read_response(reader: &mut BufReader<std::process::ChildStdout>) -> serde_json::Value {
        loop {
            let mut line = String::new();
            let bytes = reader.read_line(&mut line).expect("Failed to read line");
            if bytes == 0 {

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.