agents-learning
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| ROLE | Yes | MCP server role, supplied via --role; use worker or admin. | |
| ALP_HOME | No | Optional local data/knowledge directory. Defaults to %LOCALAPPDATA%\AgentLearning on Windows or ~/.local/share/agent-learning on POSIX. | |
| WORKSPACE | Yes | Absolute path to the project workspace, supplied via --workspace before the serve subcommand. | |
| ALP_CONFIG | No | Optional path to a custom JSON configuration file. | |
| ALP_RUNTIME_STORE | No | Optional path for fixed runtime releases. | |
| ALP_CONTEXT_HANDLE | No | Optional context handle used to bind an MCP server to a specific session. |
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 | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| candidate_reference_readB | 读取已允许参考的未发布候选；不是验证结论或执行授权。按注入卡的版本分页，状态或来源变化时停止旧版读取。 |
| candidate_reference_feedbackB | 报告已展示候选不适用或有疑问，暂停其自动参考；这是reported反馈，不证明已证伪，不发布或重试管理员。 |
| knowledge_searchA | 检索本项目正式知识的有限预览；完整正文使用knowledge_read。预算是JSON正文字符数，默认6000。 |
| knowledge_readB | 分页读取当前有效正文、适用条件或限制；section默认content，续页保持section并传next_offset与expected_revision。fields_truncated中的条件用对应section续读；默认12000个JSON正文字符。 |
| work_memory_searchB | 检索本项目已观察的工作历史索引；不是正式解法或执行授权。可按目标文件筛选，默认3000个JSON正文字符，不读取项目文件。 |
| work_memory_readA | 按需分页读取工作观察；续页保持section与expected_revision。同一上下文已完整返回的版本默认给简短回执；reread=true开启新轮，续页无需重复该标记。文件未完整核验时返回正文；不会恢复截断原文。 |
| session_contextB | 读取已验证的宿主会话与任务版本；未绑定时只返回项目状态，不推断其他会话。 |
| task_updateC | 保存当前明确的任务范围和状态；不增加执行权限。 |
| case_feedbackC | 补充专业问题、用户纠正或关键依据；常规工具事件由Hook采集。 |
| knowledge_useC | 简短声明实际采用的知识修订，可关联本任务工具操作；声明和工具观察不证明因果或用户验收。 |
| learning_statusA | 查看本项目状态；调试用view=summary的有界摘要，省略view或full保留完整状态。 |
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 11 tools
The search/read pairs for knowledge, work memory, and candidate references are distinguishable, but the boundaries between formal knowledge vs. work history vs. candidate references remain subtle for an agent. Two separate feedback tools (case_feedback and candidate_reference_feedback) also overlap in purpose, increasing misselection risk.
All names use snake_case and generally follow a resource_action pattern (knowledge_read, work_memory_search, candidate_reference_feedback, task_update). A few are resource_noun (learning_status, session_context), but the convention remains predictable overall.
Eleven tools is well within the appropriate range for a focused learning/reference server. Each tool maps to a distinct operation: status, context, task scope, knowledge access, work-memory access, candidate access, and feedback.
The surface covers reading, searching, using knowledge, and reporting feedback, but there are no tools to create, update, or delete formal knowledge or candidates. Reliance on external hooks for writes leaves notable lifecycle gaps for a learning system.