agent-memory-router
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| PYTHONUTF8 | No | Optional Python UTF-8 mode setting included in the generated .mcp.json env. | 1 |
| PROJECT_PATH | Yes | Absolute path to the project. Passed as --project_path to mcp_server.py so the server opens that project's memory.db and raw_data/. | |
| PYTHONIOENCODING | No | Optional Python I/O encoding setting included in the generated .mcp.json env. | utf-8 |
| MEMORY_ROUTER_HOME | No | Optional fallback project root when --project_path is not provided. Normally not needed for MCP configuration. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| search_memoryA | 在记忆索引中检索，返回摘要列表（不含原文）。 这是访问历史记忆的第一步。不要跳过它直接猜。 Args: keyword: 关键词，匹配标题、关键词标签和记录 ID。留空则按时间倒序返回最近记录。 type: 可选，按类型过滤。常用值：decision / code_change / issue / note / reference。 limit: 最多返回条数，1–50，默认 10。 Returns: JSON：{ok, count, results:[{id, type, title, keywords, created_at, update_version}]} 拿到 id 后，如需全文再调 read_raw_data。 |
| read_raw_dataA | 按记录 ID 读取单条记忆的完整原文。 只能对 Args:
record_id: 记录 ID，形如 Returns: JSON：{ok, id, type, title, content, size_bytes, truncated, relations} |
| save_memoryA | 把一条新的记忆写入长期库。 什么时候必须调用（Rule 里也有同样的约束）：
Args:
type: 记录类型。decision / code_change / issue / note / reference。
title: 一行摘要，要能脱离原文独立读懂（例：「改用独立浏览器上下文修复跨站污染」）。
content: 完整原文。建议包含：背景、做了什么、为什么这样做、还有什么没做。
keywords: 逗号分隔的关键词，用于以后检索（例：playwright,context,假阴性）。
relates_to: 可选，关联到已有记录，格式 Returns: JSON：{ok, id, ...}。拿到的 id 可以用于后续 relations。 |
| link_memoryB | 建立两条记忆之间的有向关系。 Args:
source_id: 起点 ID，如 Returns: JSON：{ok, source_id, target_id, relation_type} |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 4 tools
The four tools target distinct phases: search (summary list), read (full content by ID), save (new record), and link (relations between records). The only minor overlap is that save_memory can optionally create relations via relates_to, which slightly blurs with link_memory, but the descriptions clarify the intended usage.
All tools use snake_case with a consistent verb_noun pattern: search_memory, read_raw_data, save_memory, link_memory. 'read_raw_data' is slightly less consistent than a hypothetical 'read_memory', but the overall convention is clear and predictable.
With 4 tools, the server provides a focused, non-redundant set that covers the core memory operations. This is well within the ideal range, and each tool has a clear, distinct purpose.
The set covers create, read, search, and link, but lacks update and delete operations for existing memories. While the supersedes relation may mitigate some needs, there is no direct way to modify or remove a record, which is a notable gap.