Loading ToolBench

Reading the current corpus…

edgebase MCP server · F · ToolBench

edgebase

Git-native preflight context for coding agents. Provides MCP tools for context capsules, goal contracts, change radius analysis, and checkpoint management.

2026-07-28+STDIO11 tools

36/100

Definition Quality43

Protocol Readiness35

Supportability26

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

Edgebase tools have clear, domain-specific names (edgebase_context, edgebase_checkpoint) and reasonable descriptions (50-150 chars). However, critical gaps undermine production readiness: (1) Most parameters lack type definitions in visible schemas, 'changed_files' is declared as array but many params show no type info; (2) Parameter descriptions are minimal or absent, 'budget' has a description but 'task', 'goal', 'target' lack detail on format/constraints; (3) No output schemas documented, tools return 'compact context capsule' or 'Goal Capsule' but structure is opaque to LLMs; (4) Error handling is absent, no guidance on what happens if indexing fails, checkpoint is stale, or git operations error; (5) No tool annotations (readOnlyHint/destructiveHint) despite clear risk labels in metadata. Tools are functional but lack LLM-optimization patterns.

Tools (11)

edgebase_checkpointwritesource verified57/100

Save an Edgebase context checkpoint for compaction or handoff.

edgebase_contextread onlysource verified57/100

Return a compact context capsule for a task.

edgebase_finishwrite50/100

Write the final Patch Passport for the current work.

edgebase_fork_planwritesource verified57/100

Create a git worktree plan from an Edgebase checkpoint.

edgebase_goalread onlysource verified57/100

Return an executable Goal Capsule and pre-edit Work Contract.

edgebase_indexwrite50/100

Build or refresh the local Edgebase graph index.

edgebase_passportread only50/100

Return a Patch Passport for the current working tree.

edgebase_radiusread onlysource verified53/100

Issues (5)

CRITICALresponse-shaper ↗

No output schemas documented. Tools return opaque 'capsules' and 'passports' but LLMs cannot see field structure, types, or chaining IDs. Forces agents to guess what data is available.

edgebase_contextedgebase_goaledgebase_checkpointedgebase_resumeedgebase_passportedgebase_finish

HIGHtool-description ↗

Parameter descriptions are minimal or missing. 'task', 'goal', 'target', 'message' lack format/constraint guidance. LLMs cannot infer whether to pass file paths, natural language, or structured text.

edgebase_contextedgebase_goaledgebase_radiusedgebase_checkpointedgebase_fork_plan

HIGHrecovery-guide ↗

No error handling or recovery guidance. Tools do not document what happens on git failure, stale checkpoint, missing index, or invalid input. Agents have no path forward on failure.

edgebase_indexedgebase_resumeedgebase_fork_plan

Recommendations

Document output schemas for all tools. For edgebase_context, edgebase_goal, edgebase_checkpoint, specify returned fields (e.g. 'context_id', 'token_count', 'file_list', 'symbols', 'edges'). Include field types and whether they are chaining IDs for downstream tools.
Expand parameter descriptions to 50-150 chars with format/constraint guidance. E.g. 'task: Natural language coding goal (e.g., "Add user authentication"). Max 500 chars.' for edgebase_context; 'target: File paths (repo-relative) or plan text. Array of strings.' for edgebase_radius.
Add error handling and recovery guidance. Document failure modes: 'If index is stale, call edgebase_index(--changed) first.' 'If checkpoint not found, call edgebase_resume() with no snapshot_id for latest.' 'If git operation fails, verify repo state and retry.'
Expose tool annotations via MCP protocol. Mark edgebase_checkpoint, edgebase_fork_plan, edgebase_finish, edgebase_index with destructiveHint=true. Mark read-only tools with readOnlyHint=true. Mark idempotent tools with idempotentHint=true.
Add numeric bounds to 'budget' parameter across all tools. E.g. 'budget: Token budget for context (1000-4000, default 1200). Larger budgets produce more detailed analysis but consume more tokens.' Prevents LLMs from passing absurd values.

Score history

Overall score trend

First recorded score · v2 rubric

36/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	F	36	2026-07-28+	v2