Loading ToolBench

Reading the current corpus…

continuity-v2 MCP server · D · ToolBench

continuity-v2

Search and recall across every Claude Code session. Exposes tools backed by FTS5 index, TEMPORAL edge graph, and semantic embeddings for session search, similarity matching, and narrative thread recall.

2026-07-28+STDIO4 tools

56/100

Definition Quality70

Protocol Readiness50

Supportability36

Spec posture

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

Static source inference · medium confidence

Current-spec patterns detected

Summary

Four tools with complete input schemas and descriptions. Naming follows verb_noun pattern (search_, recall_, recent_, index_). Descriptions are detailed (194 - 280 chars, within baseline 34 - 392 range) and explain WHAT, WHEN, and RETURN. However, output schemas are entirely undocumented, LLMs cannot predict response structure. Parameters lack type constraints (enums, ranges); descriptions state 'optional' but no defaults documented. Error handling is minimal: search_sessions returns plain text on FTS syntax error, but no guidance on recovery. No tool annotations (readOnlyHint, idempotentHint). Composition is sound: each tool has one responsibility, and search/recall/recent form a coherent chain.

Tools (4)

index_statsread onlysource verified58/100

Quick health check of the index. Use this to verify the DB is fresh.

recall_sessionread onlysource verified73/100

Replay a session's turns. Accepts full id or unique prefix.

recent_sessionsread onlysource verified73/100

List the N most recent sessions across Claude Code and claude.ai chat.

search_sessionsread onlysource verified77/100

Full-text search across Claude Code sessions AND claude.ai chat conversations. Uses SQLite FTS5. Supports AND/OR/NOT, quoted phrases, prefix* matching. Hyphenated or numeric tokens MUST be double-quoted (e.g. '"gold-402"').

Issues (5)

HIGHresponse-shaper ↗

Output schemas completely undocumented. search_sessions returns plain-text formatted results; recall_session returns header + turns; recent_sessions returns plain-text list. LLMs cannot predict structure or extract fields for downstream chaining.

search_sessionsrecall_sessionrecent_sessionsindex_stats

HIGHconstrained-input ↗

Parameters lack type constraints. 'source' accepts 'code' or 'chat' but no enum declared. 'limit' and 'n' are unbounded integers, no min/max specified. LLMs may pass invalid values (negative limits, source='invalid').

search_sessionsrecent_sessions

MEDIUMrecovery-guide ↗

Error handling lacks recovery guidance. search_sessions returns 'FTS5 syntax error: {e}' with a hint about double-quotes, but no actionable next step. recall_session returns 'No session matching: {session_id}' with no suggestion to call search_sessions or recent_sessions.

search_sessionsrecall_session

Recommendations

Document output schemas for all tools. Define structured return types: search_sessions → {matches: [{timestamp, role, project, title, session_id, turn_idx, snippet}], total_count}. recall_session → {header: {title, session_id, project, cwd, started_at, ended_at, turn_count}, turns: [{turn_idx, timestamp, role, text}]}. recent_sessions → {sessions: [{id, title, project, started_at, turn_count, source}], total_count}. index_stats → {table_counts: {sessions, turns}, last_indexed_at, status}.
Add enum constraints to parameters. source: enum=['code', 'chat']. Add min/max to numeric params: limit: integer, minimum=1, maximum=100; n: integer, minimum=1, maximum=100; idx_from/idx_to: integer, minimum=0.
Enhance error messages with recovery guidance. search_sessions FTS error: 'FTS5 syntax error: {e}. Hyphens and numbers must be double-quoted (e.g. "gold-402"). Try: search_sessions(query="your-term", limit=10).' recall_session not found: 'No session matching "{session_id}". Try recent_sessions(n=20) to list available sessions, or search_sessions(query="...") to find by content.'
Add tool annotations to FastMCP tool decorators. Use @mcp.tool(readOnlyHint=True) for all four tools to signal safety to agents.
Expand index_stats description to 100+ chars explaining purpose and output. E.g., 'Returns database health metrics: session count, turn count, last index update timestamp, and status. Call this before search_sessions to verify the index is current and not stale.'

Score history

Overall score trend

First recorded score · v2 rubric

56/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	56	2026-07-28+	v2