deepseek-search
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| KIMI_CODE_HOME | No | Home directory from which config.toml is read to resolve the [providers.deepseek] api_key credential. | ~/.kimi-code |
| DEEPSEEK_API_KEY | No | Second-priority credential used to authenticate against the DeepSeek Search/Fetch API. Used when DEEPSEEK_SEARCH_API_KEY is not set. | |
| DEEPSEEK_SEARCH_COUNT | No | Default number of results when the caller does not pass count. | 10 |
| DEEPSEEK_SEARCH_API_KEY | No | Highest-priority credential used to authenticate against the DeepSeek Search/Fetch API. Resolved first (before DEEPSEEK_API_KEY and the [providers.deepseek] api_key in $KIMI_CODE_HOME/config.toml). | |
| DEEPSEEK_SEARCH_BASE_URL | No | API base URL; /v1/tools/search and /v1/tools/fetch are appended automatically. | https://api.deepseek.com |
| DEEPSEEK_SEARCH_MAX_RETRIES | No | Number of retries after the first attempt. | 2 |
| DEEPSEEK_SEARCH_RETRY_DELAY_MS | No | Base delay for exponential backoff. | 800 |
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 |
|---|---|
| web_searchA | Search the live web and return structured sources (title, url, snippet, summary, publish date, authority). Use it for recent events, facts you cannot confirm from memory, or anything past your training cutoff. The query accepts site:, -site:, -, "exact phrase", lang:, after: and before: operators. Follow up with web_fetch to read a page in full. |
| web_fetchA | Fetch one web page and return its title and Markdown body. Use it on a url returned by web_search when the summary is not enough. Fails when the page is unreachable, blocked by the site, or has no extractable body. |
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 2 tools
The two tools split cleanly along the search/read axis: web_search discovers sources, web_fetch reads a single page in full. The descriptions explicitly cross-reference each other ('Follow up with web_fetch', 'a url returned by web_search'), leaving no realistic misselection risk.
Both names use the same web_<action> pattern (web_search, web_fetch), same style, same prefix, unambiguous verbs. Fully predictable.
Two tools is on the thin side per the general guidance, but for a search server this is exactly the minimal viable pair—search then fetch—and each tool clearly earns its place with no filler.
Search plus full-page fetch covers the core discover-then-read lifecycle for a web search domain, with operators documented for refinement. Minor gaps exist (e.g. no batch/multi-URL fetch or crawl), but agents can work around them by calling web_fetch repeatedly.