Loading ToolBench
ProactiveAgent MCP Server: proactive memory + suggestions as pluggable MCP tools/resources/prompts for any MCP-capable agent (Claude Code, Kimi Code, Cline, Cursor). Exposes long-term memory capture/recall, behavior corrections, persona management, and proactive suggestions.
ProactiveAgent MCP exposes 20 tools with generally clear naming (verb_noun pattern) and detailed descriptions (avg 180+ chars). However, critical gaps exist: input schemas are not visible in the provided source code for most tools, only tool names, descriptions, and parameter lists are shown. The memory_capture tool shows a well-structured schema in the description (content, type, priority, scope with enums), but actual JSON Schema registration is not visible. Parameter descriptions are present and specific (e.g., 'Memory content. Keep negations intact...'), meeting the 10 - 1024 char baseline. Error handling is minimal, no recovery guidance or error classification visible. Output schemas are undocumented.
Get a specific action card by ID (unified ActionCard protocol view). Returns full card details including actions and metadata.
List all action cards (unified ActionCard protocol view). Returns card ID, type, title, and metadata.
Confirm a pending behavior correction candidate (enters active corrections; also persists as long-term memory).
Reject a pending behavior correction candidate (does not persist).
Generate a daily review summary (work done, patterns, suggestions for tomorrow). Synthesizes memories, suggestions, and activity from the current day.
Explicitly store a long-term memory (takes effect immediately, enters recall). Call immediately when the user states an explicit preference/fact/decision/correction (e.g. "以后X", "我喜欢X", "记住X", "不要X") — do not wait for wrap-up. Keep content concise, self-contained, and independently understandable. Types: fact / preference / correction / sop / todo_context / event.
Input schemas not visible in source code. Tool definitions show parameter names and descriptions in prose, but no explicit JSON Schema registration (type, minLength, maxLength, enum, etc.) is visible in the provided code.
Output schemas undocumented. No tool description specifies what fields are returned, their types, or structure. LLMs cannot plan downstream tool calls or extract data reliably without knowing response shape.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2025-06-18+ | v2 |
Confirm a pending memory (enters recall; correction/preference/sop types also refresh the user persona). Call when the user approves an item listed by memory_pending.
Feed recent conversation messages to the engine for automatic memory extraction (LLM or rule mode). Call it yourself at conversation wrap-up or when enough dialogue has accumulated — do not wait for a host hook. Extracted items default to pending (anti-poisoning), confirm via memory_pending + memory_confirm. Falls back to rule mode (zero external calls) when LLM is not configured.
List automatically extracted but unconfirmed items (anti-poisoning: only enter recall after user confirmation). Two kinds: 1. Pending memories (atom): handle with memory_confirm / memory_reject 2. Pending corrections: handle with correction_confirm / correction_reject. Call after memory_extract / memory_capture to surface what is waiting for user confirmation, and remind the user.
Search long-term memories by keywords (keyword + embedding hybrid; falls back to keyword when embedding is unavailable). Returns matching entries with type, importance, and similarity. Call at the start of a new session/task (query = current topic) to restore context; also call whenever the user references past work, project history, or personal preferences.
Reject and delete a pending memory (incorrect extraction or poisoned content).
View ProactiveAgent long-term memory statistics: memory count, type distribution, scene count, pending corrections.
Generate an onboarding guide for new users (explains ProactiveAgent capabilities, memory types, suggestion kinds, and best practices).
Retrieve the current user persona summary (synthesized from confirmed preference/correction/sop memories). Returns a concise profile of user preferences, work style, and behavioral patterns.
Explicitly save/update the user persona summary (overrides auto-synthesis). Useful when you want to manually refine the persona profile.
Get a summary of the current scene/context (synthesized from recent memories and suggestions). Returns the active work context, hot topics, and relevant background.
Accept a suggestion (marks as accepted, executes associated action if available). Call when the user approves a suggestion from suggest_list.
Ignore a suggestion (marks as ignored, does not execute). Call when the user dismisses a suggestion from suggest_list.
List all pending suggestions (not yet accepted/ignored). Returns suggestion ID, kind, title, reason, and action card.
Trigger proactive suggestion evaluation immediately (analyzes current context and memory to generate suggestions). Returns newly generated suggestions.
No error handling guidance. Tool descriptions do not explain what errors can occur, how to recover, or whether operations are retryable. Agents lack direction when calls fail.
Destructive operations lack confirmation step. memory_reject and correction_reject delete data without dry-run or confirmation. Agents can accidentally destroy memories.
Tool composition risk: memory_extract, memory_confirm, and memory_pending form a multi-step workflow. If memory_extract fails silently or returns ambiguous results, the agent may confirm poisoned data. No validation or anti-poisoning mechanism visible in descriptions.