Remote MCP server on Cloudflare Workers exposing context-keeper's rationale store over Streamable HTTP.
Server has 19 tools with consistent naming (verb_noun pattern) and documented schemas. Most tools have descriptions (10-200 chars) and input parameters with types. However, descriptions are often terse (under 50 chars), parameter descriptions lack detail on constraints/formats, and output schemas are not documented. No error handling guidance visible. Tool composition is reasonable but some tools overlap (config/set_config/get_config could consolidate). Security: no visible permission gates on destructive tools (prune_stale, deprecate_entry). Baseline: 549 production tools average 194 chars for descriptions; this server averages ~80 chars.
Get or set configuration values for a project.
Mark an entry as deprecated, optionally superseding it with another entry.
Render the project's entries as a DECISIONS.md-style Markdown document. Excludes deprecated entries unless include_deprecated is set.
Get a configuration value for a project.
Get the context for a project, including active entries and their relationships.
Get a summary of a project including counts of entries by kind and status.
Import entries from an external source into a project.
Output schemas not documented. Tools return data but LLMs cannot plan downstream calls without knowing response structure (e.g., what fields does get_context return? Does query_entries return pagination cursors?). Baseline: 100% of A+ tools document return types.
Descriptions are terse (avg 55 chars vs baseline 194 chars). Missing WHEN to use, prerequisites, and dependencies. E.g., 'Record a decision entry' does not explain when to call record_decision vs record_entry, or what fields are required.
Parameter descriptions lack constraint details. E.g., 'Configuration key' does not specify format, length, allowed characters, or examples. 'Entry payload' is opaque, what fields does it require for each kind? Baseline: descriptions should state format, range, and allowed values.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
List all projects in the store.
Delete deprecated entries older than a cutoff. Defaults to a dry run that only reports what would be removed; pass dry_run=false to actually delete.
Query entries in a project with filtering and search.
Record a constraint entry.
Record a decision entry.
Record a new entry (decision, constraint, or pipeline) in the project.
Record a pipeline entry.
Reload and validate all constraints in a project.
Set a configuration value for a project.
Update an existing entry.
Insert or update entries in a project.
Flag active entries that are missing rationale-bearing fields (decision.why_chosen, constraint.reason, pipeline.purpose) or their core field.
No error handling guidance. Tools like prune_stale (DESTRUCTIVE) and deprecate_entry (WRITE) have no documented recovery paths. If prune_stale fails, what should the LLM do? Retry? Ask the user? No guidance.
Destructive tool (prune_stale) lacks confirmation/dry-run safeguard in schema. While dry_run parameter exists, no explicit confirmation step or permission gate visible. Agents can delete data without explicit approval.