Loading ToolBench

Reading the current corpus…

Agentmemory MCP server · D · ToolBench

← Browse

Agentmemory

github.com/Obidel/Agentmemory

Visual memory manager for AI agents. MCP-native, MIT-licensed, free forever.

2026-07-28+STDIO10 tools

58/100

Definition Quality78

Protocol Readiness42

Supportability34

Spec posture

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

Static source inference · medium confidence · detected: stateful session

Deprecated protocol patterns detected

Summary

AgentMemory has well-structured tool definitions with clear naming (verb_noun pattern), comprehensive schemas with type constraints, and good descriptions. However, parameter descriptions lack depth in several tools, output schemas are not explicitly documented, and error handling guidance is minimal. The server demonstrates solid foundational quality but falls short of production-grade polish. All 10 tools have explicit registration with schemas visible in mcp/server.ts.

Tools (10)

add_memorywritesource verified87/100

Save a new memory rule/fact about the project. Use this whenever the user states a preference, decision, constraint, or piece of context that should persist across sessions.

backfill_embeddingswriteauthsource verified80/100

Generate pgvector embeddings for memories that have none. Only useful on the Supabase backend with HUGGINGFACE_API_KEY set. Idempotent and rate-limited: call repeatedly until remaining=0. Each call costs ceil(N/32) embedding API requests.

delete_memorydestructivesource verified78/100

Delete a memory by ID.

find_similarread onlysource verified75/100

Find memories semantically similar to a given text or to an existing memory. Useful for deduplication before adding new rules.

get_project_contextread onlysource verified75/100

Get all rules for the current active project, formatted as a markdown document. Use this at the start of a task to load project context.

import_jsonlwritesource verified73/100

Import a Claude Code / OpenCode / Codex JSONL session log. Parses user messages, applies heuristic category detection, and creates memories. Use this to backfill memory from past sessions.

Issues (5)

HIGHresponse-shaper ↗

Output schemas not documented. Tools return Memory objects or lists, but LLMs cannot see the structure of returned fields (id, content, category, tags, importance, project_name, source, etc.). This forces LLMs to guess field names for downstream operations.

add_memorysearch_memorieslist_memoriesfind_similarget_project_context

HIGHconstrained-input ↗

Parameter descriptions lack actionable constraints. 'query' in search_memories has no guidance on format, length, or expected content. 'limit' lacks min/max rationale. 'importance' in add_memory has no explanation of what 1-5 means semantically.

search_memoriesfind_similaradd_memoryimport_jsonl

HIGHrecovery-guide ↗

Error handling lacks recovery guidance. No tool description explains what to do if a memory is not found, if a project doesn't exist, or if import_jsonl fails to parse. LLMs receive no actionable next steps.

delete_memoryswitch_project

Recommendations

Document output schemas for all tools. For add_memory, search_memories, list_memories, and find_similar, explicitly define the Memory object structure: {id: string, content: string, category: enum, tags: string[], importance: number, project_name: string, source: string, created_at?: ISO8601, updated_at?: ISO8601}. For find_similar, add a 'score' field (0-1 similarity).
Enhance parameter descriptions with constraints and rationale. E.g., 'query (string, 1-500 chars): keyword search term. Supports partial matches. Use search_memories for keyword lookup; use find_similar for semantic matching.' Add to importance: 'Importance level 1-5: 1=nice-to-know, 3=default/standard, 5=critical constraint.'
Add error recovery guidance to tool descriptions. E.g., delete_memory: 'Deletes a memory by ID. If memory not found, returns error with available memory IDs. Use list_memories to find the correct ID.' import_jsonl: 'If parsing fails, returns error with line number and reason. Ensure .jsonl format: one JSON object per line.'
Implement confirmation pattern for destructive tools. Add an optional 'confirm' boolean parameter to delete_memory and import_jsonl. When false (default), return a summary of what would happen. When true, execute. This prevents accidental data loss.
Clarify tool selection in descriptions. Rewrite list_memories: 'List all memories in a project. Use this to browse all rules at once. For keyword search, use search_memories. For semantic similarity, use find_similar.' Add similar guidance to find_similar and search_memories to disambiguate.

Spec posture evidence

Inferred effective spec: 2026-07-28+.

Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained

Score history

Overall score trend

First recorded score · v2 rubric

58/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	58	2026-07-28+	v2