trend-radar-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| PYTHONIOENCODING | No | Python IO encoding; set to utf-8 as shown in the example MCP client configurations. | |
| PYTHONUNBUFFERED | No | Set to '1' to run Python unbuffered, as shown in the example MCP client configuration. | |
| TREND_RADAR_HOME | No | Directory where state and cache are stored (default is ~/.trend-radar). Used to override the default location for state.db, runs, and cache. |
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
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| radar_scanA | 采集 + 评分 + 分桶,一次调用给出可写简报的材料。 scope 自动判定:不给 domain 时按 query 判三类;判不进(或跨类)→ broad 全量 42 源。显式 domain 永远优先。返回 scope_source 说明是 auto / explicit / auto-broad。 返回 digest(每桶 top N:id/标题/truth·comm/来源/日期/new),并对 core、watch 的前 硬约束:truth/comm 由引擎算,0.6 是闸门,禁止自己估;borderline=true 表示分数在 0.55–0.65,必须复核后再引用;calibration 桶只校准传播度,永不作内容源。 |
| radar_packA | 按 id / bucket 取正文与细节。只在 digest 不够写的时候用。 硬约束:ids ≤ 12;limit ≤ 20;默认只给 scores+flags,要正文必须显式 fields=['summary'];摘要默认 240 字。 |
| radar_potentialA | 潜力新闻(opt-in):用现有扫描 + 历史 run 的「新增独立源」做数值预排,返回 8–12 条候选,交给模型判断谁最可能成为今天的爆点。 不传这个工具时 radar_scan 的返回完全不变;默认 5 分钟内直接复用上一次 run,不重复抓取。适合凌晨热度还没起来、按「已传播」排不出东西的时候。 返回 candidates(score/signals/why)与 meta(degraded/caveat)。caveat 固定为「未验证」,不要写成已确认新闻;不要与 core/watch 混排。 |
| radar_checkA | 发布前自检:关联度(方向性覆盖)+ 补搜条目打分,一次调用。 硬约束:量的是准备发布的条目(bucket=core 或显式 entries),不要传全池;关键词 ≤ 8;补搜条目 ≤ 10 且必须带 url;<0.6 必须重查后再写。 |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| source registry | 42 个源、三路分类、13 条已知死路 |
| state & source health | 状态库统计 + 连续空源黑名单 |
| recent runs | 最近的扫描记录(run_id / 时间 / 统计) |
TDQS
Scored across 4 tools
Each tool has a fairly distinct role: radar_scan collects and buckets, radar_pack fetches full text, radar_check validates pre-publish entries, and radar_potential ranks emerging candidates. The main overlap is between scan and potential, both returning candidate sets, but descriptions clarify the opt-in, pre-heat use case for potential.
All four tools use a consistent radar_ prefix followed by a short snake_case token. The pattern is predictable and readable throughout.
Four tools is slightly thin but well-scoped for a focused scan-pack-check-potential pipeline. Each tool has a clear job and there is no obvious bloat.
The surface covers the core workflow: scanning, deepening with full text, pre-publish validation, and opt-in potential detection. Minor gaps exist, such as no explicit history/run lookup or configuration tool, but agents can work around these.