zhiyan-detective
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| DETECTIVE_HOME | No | Default directory for the usage ledger and cache. Defaults to ~/.zhiyan-detective. | |
| DETECTIVE_LEDGER | No | Path to the JSON usage ledger. Every real search is recorded before the request; requests stop when the cap is exceeded. Point multiple entry points to the same file to share one ledger. | |
| DETECTIVE_DAY_CAP | No | Daily search cap. Can only be lowered. Default is 1200. | |
| DETECTIVE_CACHE_DIR | No | Directory for cached search results. The same query is reused within 24 hours without spending quota. Defaults to $DETECTIVE_HOME/cache. | |
| DETECTIVE_CASES_DIR | No | Optional directory where closed case files are stored as JSON, up to 500, so report links remain available after restart. Text recognized from screenshots is not stored. | |
| DETECTIVE_MONTH_CAP | No | Monthly search cap. Can only be lowered. Default is 5000. | |
| DETECTIVE_WSA_KEY_FILE | No | Path to a private file containing a line like TENCENTCLOUD_WSA_APIKEY=<your key>; alternative to setting TENCENTCLOUD_WSA_APIKEY directly. | |
| TENCENTCLOUD_WSA_APIKEY | No | Tencent Yuanbao Search API key used by the MCP server. | |
| DETECTIVE_MCP_DAILY_CASES | No | Maximum number of cases per process per day for the MCP server. Can only be lowered. Default is 60. |
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
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| verify_claimA | 查证一条 AI 产品消息(发布日期、是否开源、哪家发布、是否发布)。返回结论、官网/官方仓库/媒体证据、山寨官网提醒和每句都带出处的公众号查证稿。每条最多 4 次元宝搜索。 |
| verify_digestA | 一次查证最多 5 条 AI 消息,返回每条结论和一篇统一编号出处的快讯合集(Markdown),可直接改写成公众号文章。 |
| find_workA | 凭一句台词、歌词或剧情找剧和歌:两个独立来源对上才结案,对不上返回假说和一个追问,附正版链接。 |
| answer_follow_upB | 回答 find_work 的追问(例如角色名或年份);每个案子只能追问一次。 |
| usageA | 查看元宝搜索额度:本月和今天已用次数、上限,以及这个进程今天已查的案数。 |
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
verify_claim and verify_digest overlap in purpose but are clearly separated by single-claim vs. batch (up to 5) scope, and find_work/answer_follow_up are linked but distinct (initial search vs. one allowed follow-up). usage is unmistakably separate. Only minor confusion possible between the two verify tools when a user has a single item that could also go into a digest.
Names are consistently snake_case and mostly follow verb_noun: verify_claim, verify_digest, find_work, answer_follow_up. The lone exception is usage, a bare noun rather than an action verb, which is a minor stylistic deviation.
Five tools is well within the ideal 3-15 range, and each earns its place: single verification, batch verification, discovery search, its follow-up path, and quota inspection. Nothing feels redundant or padded.
Core workflows are covered: verify one claim, verify a batch and produce a digest, find a work with a bounded follow-up, and check quota. Minor gaps exist—no way to retrieve prior verification results or case history, and find_work has no batch variant—but agents can work around these.