context-saver-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| CONTEXT_SAVER_CONFIG | No | Optional path to the Context-Saver JSON config file. Search order is --config <path>, $CONTEXT_SAVER_CONFIG, ./context-saver.config.json, then ~/.context-saver/config.json. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
Context-Saver is a lazy MCP gateway: only the 6 cs_* meta-tools are always loaded; every other tool lives on downstream servers. Workflow for each task: (1) cs_discover or cs_plan to find the right tools (cheap, no schemas); (2) cs_load exactly the tools you need — they become normal callable tools and their schemas are echoed in the result; (3) call the loaded tools directly, or use cs_call for one-off calls (auto-loads, no schema needed); (4) cs_unload when done to reclaim context; (5) cs_status for diagnostics and token savings. Never guess downstream tool names — always discover or plan first.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| cs_discoverA | List or search every tool available through the gateway across all configured downstream MCP servers. Returns id, source server and a one-line description — no schemas, minimal tokens. Always do this before cs_load. |
| cs_planA | Inspect the user's task and recommend the minimal set of downstream tools to load. Scores the task against the whole catalog and returns ranked matches with reasons. Use when unsure which tools a task needs. |
| cs_loadA | Load specific downstream tools into this session: their full input schemas are returned here and the tools become directly callable (the client also receives tools/list_changed). Accepts exact ids ("server__tool"), globs ("github:*") or bare tool names. |
| cs_unloadA | Unload previously loaded downstream tools. Omit "tools" to unload everything loaded this session. |
| cs_callA | Invoke any catalog tool by id right now, without loading its schema into the session (arguments pass through verbatim). Prefer cs_load when you need the input schema; use cs_call for one-off calls. |
| cs_statusA | Show downstream servers and connection states, loaded tools, and live token-savings accounting for this session (context used vs. direct connections). |
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 6 tools
Each tool has a distinct role in the gateway lifecycle: discovery, planning, loading, unloading, direct invocation, and status. cs_discover and cs_plan are related but clearly differentiated by catalog listing versus task-specific recommendation.
All tool names follow a consistent cs_ prefix with a clear verb-based convention (discover, plan, load, unload, call, status). The naming is predictable and easy to scan.
Six tools cleanly cover the full lifecycle of discovering, selecting, loading, calling, unloading, and monitoring downstream tools. The set is well-scoped with no redundant entries.
The surface covers discovery, planning, schema loading, direct invocation, unloading, and status/accounting. There are no obvious dead ends for managing context-saving gateway workflows.