Exa Skills

Agent Skills for Exa web search — without an MCP server.

A collection of standalone Agent Skills that give any SKILL.md-compatible coding agent (Claude Code, Codex, Cursor, OpenCode, Gemini CLI, …) first-class access to Exa's semantic search, content fetch, and answer APIs. Each skill calls the Exa REST API through a small, zero-dependency Python script — the only thing you need is an EXA_API_KEY.

Why not just the MCP server? MCP is great, but it adds a process to run, a connection to manage, and a tool surface to wire up. These skills need nothing but Python 3.8+ and a key — they're portable, inspectable, and work the same whether your agent supports MCP or not.

---

The skills

Skill	Use it for	Default category
exa-native-base	The foundation — query-writing, the shared client, subagent orchestration. Read first.	—
exa-web-search	General web search — the search_web analog. Start here for open-ended lookups.	(any)
exa-company-research	Companies, competitors, funding, market maps	company
exa-people-research	People by role / company / location (LinkedIn-weighted)	linkedin profile
exa-lead-generation	Prospect lists: companies → decision-makers → enrichment	company
exa-research-paper-search	Academic papers, surveys, arXiv preprints	research paper
exa-code-context	Code examples, API usage, docs, GitHub repos, error fixes	(web) / github
exa-financial-report-search	SEC filings (10-K/10-Q/S-1), earnings, annual reports	financial report
exa-personal-site-search	Personal blogs, portfolios, homepages	personal site
exa-x-search	X / Twitter posts (limited coverage — see the skill's notes)	(x.com/twitter.com)

---

Quick start

# 1. Clone
git clone https://github.com/CodeAlive-AI/exa-skills.git
cd exa-skills

# 2. Set your key (get one at https://dashboard.exa.ai/api-keys)
export EXA_API_KEY=your-key
#   …or:  cp .env.example .env  &&  edit .env

# 3. Run any skill's script directly
python exa-company-research/scripts/company_research.py "AI infrastructure startups" -n 10
python exa-research-paper-search/scripts/research_paper_search.py "retrieval augmented generation survey"
python exa-native-base/scripts/exa.py answer "Who founded Exa?"

Every search prints a clean Markdown list ending with sources_reviewed: N. Add --json for raw output, --text to include page contents, -h to see all flags.

Using them as agent skills

Each exa-* folder is a standard Agent Skill (SKILL.md + scripts/). Your agent reads the SKILL.md, then runs the script.

📦 Full per-agent install guide → INSTALL.md (Claude Code, Codex, Cursor, OpenCode, Gemini CLI — plus a note on closed builders like Orchids, which can't install skills).

Just want general web search? exa-web-search is self-contained — copy that one folder and you're done:

git clone --depth 1 https://github.com/CodeAlive-AI/exa-skills /tmp/exa-skills
cp -r /tmp/exa-skills/exa-web-search ~/.claude/skills/exa-web-search   # Claude Code, user-level
export EXA_API_KEY=your-key

Want the whole collection (specialized skills + the exa-native-base foundation):

• Claude Code / Codex / OpenCode / Cursor / Gemini CLI: make this repo discoverable as skills (clone it into your agent's skills directory, or symlink the exa-* folders). Keep the folder structure intact — the specialized skills' scripts locate the shared client (_shared/exa_client.py) by walking up to the repo root.
• Set EXA_API_KEY in your environment (recommended) so it's available wherever the agent runs scripts.
• Start the agent on a research task ("research competitors to Acme", "find recent papers on X", "who leads engineering at Y") and it will pick the matching skill.

How it works

exa-skills/
├── _shared/exa_client.py        # the engine: Exa REST client + CLI (stdlib only)
├── exa-native-base/             # foundation skill: philosophy, references, generic CLI
│   ├── SKILL.md
│   ├── scripts/exa.py           # generic: search / contents / answer / similar
│   └── references/              # searching, filtering, extraction, synthesis, source-quality, full SDK spec
├── exa-company-research/
│   ├── SKILL.md                 # when-to-use + domain query patterns
│   └── scripts/company_research.py   # ~20-line preset over the engine
├── … one folder per skill …
└── tests/smoke.py               # offline checks + optional live smoke test

• _shared/exa_client.py wraps Exa's /search, /contents, /answer, and /findSimilar endpoints. It auto-loads a .env from the repo root, gives friendly errors for missing/invalid keys and rate limits, and supports categories, domain/date/text filters, and client-side domain enforcement.
• Each skill is a thin preset (default category + curated query patterns) plus a SKILL.md that teaches the agent how to use it well — including token isolation (run bulk searches in subagents, not your main context).

Notes from building this (verified against the live API, 2026-06)

• Valid category values: company, research paper, news, pdf, github, personal site, linkedin profile, financial report.
• The old tweet category was retired — exa-x-search filters to x.com/twitter.com instead, and is honest that coverage is now limited.
• type values are auto (default), fast, deep, deep-reasoning, instant.

Requirements

Python 3.8+ and an Exa API key. No third-party packages. (The Exa Python SDK, exa-py, is optional and not used.)

Acknowledgements

Query patterns and research-orchestration guidance are adapted from Exa Labs' open-source search skill (MIT). This project reimplements those ideas as standalone, script-based skills that need no MCP server. Not affiliated with or endorsed by Exa Labs.

License

MIT © CodeAlive

---

_{Maintained by CodeAlive — semantic code search and AI-powered answers across your codebase.}