Loading ToolBench
Local MCP (Model Context Protocol) server for Howdex operational memory. Provides tools for storing execution traces, learning procedures, retrieving guidance, and managing a local Codex catalogue.
Howdex MCP server has 17 tools with mixed quality. Naming is generally clear and verb-based (remember, learn, guidance, search, publish, attach, forget, stats, inspect, run, check, build, deploy). Descriptions are present for all tools and range from 50-150 chars, meeting the 10-1024 char baseline. However, input schemas vary significantly in completeness. Core tools (howdex_remember_trace, howdex_learn, howdex_guidance) have well-structured schemas with typed parameters and descriptions. Deployment tools (build_docker_image, run_database_migration, deploy_service) have empty input schemas ({}), scoring 0 for schema. Parameter descriptions are generally adequate but some lack constraint details (e.g., howdex_remember_trace 'steps' array accepts oneOf string/object without clarifying structure). Error handling and output schemas are not visible in source. Tool composition is reasonable, each tool has a single responsibility. Security concern: no evidence of permission gates or audit logging for destructive operations (howdex_forget, deploy_service).
Build the Docker image for the API service.
Check whether DATABASE_URL is available in the production environment.
Check whether the database migration file exists.
Deploy the API service.
Attach inspectable verification evidence to a procedure. A receipt is verified only when exit_code is zero and observed_signal contains expected_signal.
Publish one local learned procedure as a Codex entry. Unverified procedures publish as candidate entries.
Search a local Howdex Codex catalogue with relevance, policy, and staleness controls.
Six deployment tools (run_tests, check_DATABASE_URL, check_migration_file, build_docker_image, run_database_migration, deploy_service) have empty input schemas ({}). Per hard scoring rules, schema score must be 0 for tools with no input schema visible.
Destructive tool howdex_forget lacks permission gates, audit logging, and confirmation pattern. No evidence of access control or recovery guidance in error handling.
howdex_remember_trace 'steps' parameter accepts oneOf [string, object] without documenting the object structure. LLMs cannot infer valid object shapes, risking malformed input.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Delete a generic memory by id.
Retrieve relevant local procedures and Codex entries, then render bounded operational guidance for the current objective.
Consolidate successful episodic traces into reusable procedures.
Store a generic Howdex memory.
Store one structured execution trace as an episodic Howdex session. Use this after an agent run to preserve what was tried, what failed, and what worked.
Search generic Howdex memories.
Return local Howdex database stats.
Inspect Howdex memory before planning deployment.
Run the database migration.
Run the test suite before deployment.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields without knowing response structure.
Deployment tools lack error handling guidance. No indication of retryable vs fatal errors, or what the LLM should do on failure (e.g., 'Docker build failed: check Dockerfile syntax').