Claude Skill
telem-search is an installable Claude Agent
Skill — a SKILL.md plus
a self-contained helper script — that teaches Claude to search the web by
calling the telem-sdk Python client directly. It’s an alternative to the
MCP server for hosts that load Agent Skills natively:
Claude reads the skill’s instructions and writes/runs Python against the SDK
itself, rather than calling a separate stdio process.
Install
Section titled “Install”The skill calls telem-sdk — see the Python SDK page for the
full client API. The skill itself isn’t a published package — it’s the
skills/telem-search
directory inside the TelemSDK repo, which is private today. Contact the Telem
team for access. Once you have a checkout, run the install below from its
root — or set TELEM_REPO to point at the checkout from anywhere else.
pip install telem-sdkmkdir -p ~/.claude/skillsTELEM_REPO="${TELEM_REPO:-.}"cp -r "$TELEM_REPO/skills/telem-search" ~/.claude/skills/telem-searchThis installs telem-sdk, then copies the skill directory into
~/.claude/skills/telem-search so Claude Code picks it up in every project.
To scope it to one project instead, copy it into that project’s
.claude/skills/telem-search directory.
What the skill teaches
Section titled “What the skill teaches”Once installed, Claude has instructions for:
- A single search, and batched multi-query searches — several queries run concurrently server-side as one interaction.
- Session continuation: pass the session id back on every follow-up search serving the same user goal, and only set a goal on the first search of a new one.
- Listing and selecting providers.
All of it goes through the same Telem/AsyncTelem client documented on the
Python SDK page — the skill adds no new API surface.
Running the script standalone
Section titled “Running the script standalone”The skill ships a self-contained CLI, scripts/search.py, that works without
Claude — a plain one-shot search from a terminal. Run it from inside the
installed skill directory:
uv run --with telem-sdk python scripts/search.py "query" --goal "..."It accepts one or more queries (batched into a single interaction), plus:
| Flag | Meaning |
|---|---|
--session <id> |
Continue an existing session (omit --goal when set) |
--goal "..." |
Overall user goal — only on the first search of a new session |
--providers exa,tavily |
Comma-separated provider allow-list |
--max-len <n> |
Per-result content cap in characters (default 8000) |
It prints the session id first, then the formatted results, so a shell script
can capture the id for a follow-up call with --session.
Configuration comes from the environment, same as the SDK: TELEM_API_KEY and
TELEM_BASE_URL — see Authentication.