agentsmd-memory
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| MEMORY_FILE | No | Target file name, e.g. CLAUDE.md, GEMINI.md. Bare name only. | AGENTS.md |
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 |
|---|---|
| memory_saveA | Consider saving non-obvious project decisions or gotchas at task completion only when they prevent a likely future mistake or substantial repeated work. Skip facts cheaply discoverable from code or docs, task summaries, temporary state, and duplicates. Batch related learnings; no update is usually needed. Returns a resolved memory path, size guidance, and instructions to assess and merge with your own editing tools; never writes files or requires saving a low-value candidate. |
| memory_forgetA | Correct or remove stored facts promptly when evidence shows they are wrong, superseded, or obsolete. Describe the facts in natural language; the agent checks evidence and edits them, with no automated fuzzy matching. Returns instructions for your own editing tools; preserves unrelated content and never writes files. |
| memory_reviewA | Review and trim project memory when the user requests cleanup or a major project change makes guidance obsolete. Do not call every turn or after routine tasks. Returns the resolved file, word count, soft budget, and evidence-based cleanup instructions; never edits or deletes files itself. |
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 3 tools
memory_save (add facts), memory_forget (fix/remove wrong facts), and memory_review (bulk cleanup on request) target largely distinct actions, and the descriptions explicitly distinguish the triggers. There is mild overlap between forget and review since both edit existing content, but the conditions (evidence-based correction vs. user-requested cleaning) keep them separable.
All three tools follow an identical memory_<verb> pattern with clear, action-oriented verbs (save, forget, review). The convention is fully consistent and readable.
Three tools is on the thin side but each maps to a distinct memory lifecycle action (add, correct, clean), so none feel redundant. It sits at the low end of the ideal range but is defensible for this narrow purpose.
Save, correct/forget, and prune are covered, but there is no explicit read/list/recall tool to surface stored memory, and none of the tools actually writes files (all return instructions for the agent's own editing tools). The lifecycle is mostly present but has a notable retrieval gap.