context-guardian
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| GUARDIAN_HOST | No | Interface the running Context Guardian proxy binds; the MCP server reads the proxy's `/guardian/stats` over this host. Leave this alone unless you know what you are doing. | 127.0.0.1 |
| GUARDIAN_PORT | No | Port the running Context Guardian proxy listens on; the MCP server reaches it at `GUARDIAN_HOST`:`GUARDIAN_PORT` for `guardian_status`. | 8786 |
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
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| guardian_statusA | Context Guardian's live status from the running proxy: window and budget, compactions, tokens saved, rejected summaries, calibration. Reads GUARDIAN_HOST:GUARDIAN_PORT. |
| search_spansA | Case-insensitive search of the conversation text compaction archived. One line per hit: span file, message index and role, snippet. |
| read_spanA | The summary and original messages of one span, by run id and index (as search_spans prints them). Long spans are cut at 20000 characters. |
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
The three tools have clearly distinct purposes: guardian_status reads live proxy metrics, search_spans searches archived conversation text, and read_span fetches one specific span's content. Although search_spans and read_span both touch spans, the search-vs-read boundary is unambiguous.
All names use consistent snake_case, which is good. There's a minor deviation in verb style: search_spans and read_span are verb_noun, while guardian_status is a noun phrase (status) rather than an action, but the names remain predictable and readable.
Three tools is on the lean side, but for a focused read-only observability surface (status + archive search + span read) each tool earns its place. It's slightly thin rather than excessive.
Status, search, and span retrieval cover the main inspection workflow. Minor gaps exist, such as enumerating all runs/spans without a prior search or triggering/configuring compaction, but these are workable around for an inspection tool.