The value is in the sequence: staff it, solve it, score it, prioritise it, then find what can run at the same time — each step reading the same problem record. It plans and critiques; execution stays with whoever picks the plan up.
An MCP server that runs a structured problem-solving workflow in the client's context. You define a problem with a complexity score; it proposes a team of 3 to 12 professional roles, generates a solution, scores that solution across completeness, feasibility, quality, risk and timeline, sorts the work with an Eisenhower matrix, and analyses which tasks can run in parallel. State is held per problem id, so the later tools all work off the same case.
- A problem recorded with title, description, domain and a complexity score, then solved end to end — `create_problem`, `solve_problem`
- A recommended team sized to that complexity, drawn from 12 role types — `get_role_recommendations`
- The proposed solution scored across completeness, feasibility, quality, risk and timeline, with issues classified by severity and specific improvements returned — `check_solution`, `improve_solution`
- Tasks sorted into important/urgent quadrants for a priority order you can defend — `eisenhower_matrix_analysis`
- Dependencies between tasks mapped, and the parallel execution plan that follows from them — `analyze_task_dependencies`, `optimize_parallel_execution`
- Team state, member updates and task assignment with priority — `get_team_status`, `update_team_member`, `assign_task`
- An execution report and the history of problems handled so far — `get_execution_report`, `get_problem_history`
- Reflection records with insights and lessons, and a summary across them — `create_reflection`, `get_reflection_summary`
Node.js 18 or higher and npm 8 or higher. Installation is a clone and build — `npm install`, `npm run build` — after which the client runs `node dist/index.js` with the project directory as `cwd` and `NODE_ENV` set to `production`. `MAX_TEAM_SIZE` caps team expansion at 30 and `PARALLEL_THRESHOLD` sets how repetitive a task has to be before it is split, default 0.7. Nothing is stored outside the process: problems live in memory for the session, so the history tools cover the current run rather than an archive.