AI Workspace Kit
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| --state | No | Path to the private state directory. Defaults to ~/.local/share/ai-workspace-kit. If you use a different state folder, you must pass the same --state to artifacts and the MCP server. | |
| --memory | No | Path to the memory directory. Defaults to ~/ai-workspace-memory. If you change the location, pass the same --memory to all commands. |
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
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| workspace_recallA | Recall prior work when the user says continue a named project, where did we leave off, or 예전에/어디까지 했지/이어서 하자. The user need not say AI Workspace. Use factual topic keywords; if the topic is missing, resolve it from current conversation or ask. Search recent and refined memory, checking freshness. Only when operator-enabled, fall back to bounded redacted Cold excerpts. No full sessions or local paths returned. |
| workspace_checkpointA | Save a strict v1 confirmed_summary checkpoint locally; never raw chat. Required keys: version=1, source=chatgpt|codex|manual|aside, session, checkpoint, stamp with timezone, kind=confirmed_summary, verified=true, explicit_memory boolean, next_context short text, memory object. memory requires topic/repo slugs, business boolean, sensitive=false, and arrays context/status/decisions/todo/failed_approaches/links/completed_todo. No secrets, personal data or local paths. Stable session/checkpoint IDs ensure idempotency. Existing scheduler handles promotion; this call does not publish. |
| search_localA | Find original work documents on explicitly allowed local/SSD roots. Use factual keywords such as ExampleCo 견적서. Returns filename, full original path, modified time, topic, excerpt and artifact id. Source may be offline. Do not answer amounts/scope from snippets: select a candidate and call read_local_artifact. Session conversations use workspace_recall instead. |
| read_local_artifactA | Fresh read of an id returned by search_local, never an arbitrary path. Read-only bounded extracted text, original path and source fingerprint. Respect missing/disconnected/stale/blocked errors and pagination; quote only content actually read. PDF rendering/download is not provided by text extraction. |
| workspace_healthB | Check availability and scope without exposing local paths or historical content. |
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 5 tools
Each tool has a clearly distinct purpose: memory recall, checkpoint persistence, local document search, artifact reading, and health checking. The descriptions explicitly separate session memory (workspace_recall) from local document search (search_local) and clarify that read_local_artifact must follow search_local. This leaves little chance of misselection.
All names use snake_case, but the set mixes two prefix families (workspace_ and local/search/read) and inconsistent verb/noun structures such as workspace_recall, workspace_checkpoint, workspace_health, search_local, and read_local_artifact. The names remain readable, but the pattern is not fully predictable.
Five tools is well-scoped for a workspace memory and local-document kit. Each tool covers a distinct operation and none feels redundant or gratuitous. This is squarely within the ideal 3–15 range.
The surface covers core retrieval, persistence, local search/read, and health checking, which are the main workflows for this domain. Minor gaps remain around memory lifecycle operations such as updating, deleting, or listing checkpoints, but agents can likely work around these limitations.