context-keeper
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| CONTEXT_KEEPER_PROJECT | No | Path to the project directory (optional, server resolves automatically if omitted). | |
| CONTEXT_KEEPER_REMOTE_URL | No | URL of the remote mirror endpoint (optional, e.g. https://context-keeper-remote.<acct>.workers.dev/mcp/<AUTH_TOKEN>). | |
| CONTEXT_KEEPER_REMOTE_TIMEOUT | No | Optional per-request timeout in seconds for remote mirror (default 5). |
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 |
|---|---|
| record_entryA | Unified write tool: record a decision, constraint, or pipeline. Required fields by kind (validated server-side): decision needs summary/problem/why_chosen; constraint needs rule/reason; pipeline needs name/purpose/steps. |
| get_contextB | Retrieve relevant project context ranked by relevance within a token budget. Pass an id to fetch a single entry at full fidelity. |
| query_entriesA | Structured field filter — exact AND-combined predicates, NOT relevance search. Use when you know the field values (e.g. absolute constraints scoped to 'hooks/'). Stable ID order, no ranking; empty is a real answer. Includes superseded/deprecated unless you filter status. |
| get_project_summaryA | The single orienting call for conversation start: counts by kind and status, active constraints (compact), recent decisions, the full id list, and the human-readable summary — all in one response. |
| update_entryA | Update an existing decision, pipeline, or constraint by ID. Refreshes verified_at timestamp automatically. |
| deprecate_entryB | Mark an entry as deprecated. For decisions, optionally link to the superseding decision. |
| prune_staleA | Find entries not verified in N days. Returns them for review — does not delete. Call periodically to keep context fresh. |
| get_compaction_reportA | Check if the last compaction lost or modified any context entries. Surface discrepancies to the user before proceeding. |
| verify_qualityC | Scan entries for quality issues (legacy schema, thin reasoning, missing tags, isolated entries) and return them for enrichment via update_entry. |
| export_markdownB | Regenerate DECISIONS.md from the decisions store — a derived, read-only projection; overwrites the file whole. |
| reload_constraintsA | Re-surface the project's constraints only (not the full store) when a long session has buried the ones injected at start. Same block as session start. |
| export_snapshotA | Write the whole store to a compressed, committable snapshot at .context-keeper/memory.json.gz (next to the project) and add a .gitattributes merge=ours guard. For sharing project memory with a team via git. Committing it is opt-in. Non-destructive to the working store. |
| import_snapshotA | Import the committed .context-keeper/memory.json.gz snapshot into the working store. Non-destructive: a store that already has entries is left untouched. Runs automatically on first use when the store is empty. |
| mirrorA | Sync entries with the remote store. op='pull' merges remote entries into local (newest wins); op='backfill' pushes all local entries to remote (upsert). No-op if remote unconfigured. |
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 14 tools
Each tool has a clearly distinct purpose: creation, retrieval by relevance, structured filtering, overview, snapshot export, update, deprecation, stale pruning, compaction reporting, quality verification, markdown export, constraint reloading, snapshot import, and remote sync. No overlap in functionality.
All tool names follow a consistent verb_noun pattern using snake_case (e.g., record_entry, get_context, export_snapshot). The single verb 'mirror' is common and fits the pattern of operations.
14 tools cover the full scope of context management without bloat. Each tool addresses a specific need (CRUD, maintenance, export/import, sync) and the count is appropriate for the domain.
The tool surface covers the full lifecycle: create, read (multiple modes), update, deprecate, export/import, sync, maintenance (pruning, quality checks), and constraint management. No obvious gaps for the stated purpose of project context management.