How perftool-incubator/crucible works
Crucible is a modular performance-testing harness: users define runs, Crucible launches benchmark and tool workloads across supported endpoint types, then converts and stores results for querying and comparison. Its MCP service offers a secured, typed interface for run planning, operations, and job supervision. The README documents the harness workflow and major capabilities; the MCP and logger edges below are grounded in the supplied source excerpts and SOURCE INDEX. Several harness subsystems are represented at their documented directory-level boundary because their implementation was not sampled.
Main entry points: server.py MCP transport · bin Benchmark engine · operations.py Typed operations · crucible Crucible CLI
Components 26
Run orchestration
- Crucible CLI
bin/crucible - Run definition
- Endpoints
bin
Workload execution
- Tool collection
bin - User environments
bin/userenvs - Benchmark engine
bin - Benchmarks
config/repos.json
Results and logging
- Result processor
bin/result-processor.py - Common data model
bin/result-processor.py - OpenSearch
bin/manage_opensearch.py - Query interface
bin - Logger CLI
bin/_logger.py - Log commands
bin/_log.py - Log database
bin/_logger_lib/db.py - Log viewer
bin/_logger_lib/viewer.py
MCP service
- MCP transport
mcp/crucible_mcp/server.py - Security policy
mcp/crucible_mcp/policy.py - Typed operations
mcp/crucible_mcp/operations.py - Docs resources
mcp/crucible_mcp/documentation.py - Job supervisor
mcp/crucible_mcp/runner.py - Durable jobs
mcp/crucible_mcp/jobs.py - Job models
mcp/crucible_mcp/models.py - Audit
mcp/crucible_mcp/audit.py
Other components
- User
- Systems under test
- MCP client
Connections 26
- User → Crucible CLI: invokes
README.md - Crucible CLI → Run definition: accepts
README.md - Crucible CLI → Endpoints: selects
README.md - Endpoints → Benchmark engine: launches
README.md - Endpoints → User environments: uses image
README.md - Benchmark engine → Benchmarks: executes
README.md - Benchmark engine → Tool collection: collects data
README.md - Benchmark engine → Systems under test: tests
README.md - Result processor → Common data model: converts output
README.md - Common data model → OpenSearch: stores results
README.md - User → Query interface: queries results
README.md - Logger CLI → Log database: writes logs
bin/_logger.py - Log commands → Log database: reads logs
bin/_log.py - Log commands → Log viewer: delegates viewing
bin/_log.py - MCP client → MCP transport: sends requests
- MCP transport → Security policy: checks credentials
mcp/crucible_mcp/server.py - MCP transport → Typed operations: dispatches operations
mcp/crucible_mcp/server.py - MCP transport → Job supervisor: dispatches jobs
mcp/crucible_mcp/server.py - MCP transport → Durable jobs: uses job store
mcp/crucible_mcp/server.py - MCP transport → Audit: uses audit
mcp/crucible_mcp/server.py - MCP transport → Docs resources: serves resources
mcp/crucible_mcp/server.py - Job supervisor → Durable jobs: persists state
mcp/crucible_mcp/runner.py - Job supervisor → Typed operations: validates plans
mcp/crucible_mcp/runner.py - Durable jobs → Job models: builds records
mcp/crucible_mcp/jobs.py - Typed operations → Security policy: checks inputs
mcp/crucible_mcp/operations.py - Typed operations → Docs resources: uses catalog
mcp/crucible_mcp/operations.py