SkillFoxx

Exa Research MCP Skills

Four research skills for company, competitor, foreign trade and renewable energy market intelligence through Exa.

  • 118 ★ on the repository
  • Last commit
Skill

Extended access

Why this level

  • Company and market names go to Exa and web pages during search.
  • File mode writes many JSON, CSV and report files into the project folder.
All reasons and checks

liangdabiao/exa-research-mcp-skillAuthor:liangdabiao

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add skills/exa-research-mcp-skills

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Extended accessThe tool runs code or makes network calls and only reads external data.

  • Company and market names go to Exa and web pages during search.
  • File mode writes many JSON, CSV and report files into the project folder.

This is third-party code. Review the repository files before installing.

What it does

A set of four portable skills in the `skills/` folder: `company-research`, `foreign-trade-research`, `renewable-market-research` and `effective-harnesses`. The first collects facts about a company and its rivals: funding, news, headcount, public leadership profiles. The second builds a business report in Chinese for a country and product pair: twenty market players (seven brands, seven manufacturers, six distributors), a comparison matrix and market gaps. `renewable-market-research` runs deep renewable energy market research in file mode: search results land in JSON files, from which a project ledger and two reports, full and short, are assembled. `effective-harnesses` keeps `feature_list.json` and `agent-progress.md` so a long task survives a session change. Search runs through Exa MCP or any web search the agent has, and reports keep their sources.

Who it is for. Analysts, exporters and sales managers who gather company and market facts with sources and want raw search kept out of the main conversation.

Good fit when

  • Before a meeting you collect a company brief: news, funding rounds, headcount, leadership.
  • An exporter needs a list of twenty market players in a chosen country for one product category.
  • A team enters the renewable energy market and keeps a project ledger by stage.
  • Research stretches across several sessions and progress is kept in files.
  • Competitors are compared in a table with a source for each claim.

Not a fit when

  • The research needs closed databases that are absent from the open web and Exa.
  • Figures go into a report unchecked: the skills separate facts from inference, but a person does the verification.
  • The task is a single question that ordinary search already answers.

Example request

Research the competitors of Cursor, Windsurf and Continue, output a table and cite a source for every claim.

Limitations

The skills expect Exa MCP (`claude mcp add --transport http exa https://mcp.exa.ai/mcp`); without it the agent's own web search is used and results are weaker. Installation in the README comes down to copying folders from `skills/`, with no installer. The `renewable-market-research` mode runs multi-round search and spends many Exa requests. Part of the materials and the sample reports are in Chinese. On Windows without Bash, use PowerShell or Python.

How to disable. Remove the copied skill folders from the agent's skills directory and the Exa connection from the MCP configuration. Report files under `data/renewable-market/` stay in the project and are deleted by hand.

FAQ

Is Exa required?

No, but it is recommended: Exa MCP searches companies, news and people better. Without it the skill uses whatever web search the agent has.

How do I invoke a skill?

Name it in the request: `$company-research`, `$foreign-trade-research`, `$renewable-market-research` or `$effective-harnesses`. Then describe the country, technology or company and the report you want.

What is file mode?

Worker agents write results to `data/renewable-market/depth/*.json`, and the main agent merges them into a main JSON, CSV and reports. That keeps a large search out of the main conversation.

Do the skills work in Codex and OpenClaw?

Yes. `SKILL.md` holds only `name` and `description`, and `agents/openai.yaml` next to it helps the Codex interface. For OpenClaw, copy the folders to `~/.openclaw/skills/`.

How does the foreign trade research skill work?

`foreign-trade-research` takes a country and a product, finds twenty market players, profiles each one, builds a competitor matrix and cross-checks data across several sources. The output is a business report in Chinese.

Discussion

No questions or reviews yet.

Skills
Official

Official Exa MCP server: web search, reading pages as Markdown and multi-step research via Exa Agent

MCP serverExtended access5.2K
Official

Official Exa skills connect an agent to the Exa API: semantic search, page extraction, answers, company research and lead lists

SkillExtended access70

Three search skills: four-source search scored by intent, deep thread tracing and clean Markdown extraction from pages.

SkillExtended access438

A CLI and MCP server that sends one question to search and AI sources and saves answers, citations and cost in a run.

CLIExtended access138

A research MCP server: searches the web, SEC, courts and patents, reads whole pages and checks that cited sources really exist.

MCP serverBroad access77

Three skills and four utilities that let an agent search the web, read pages, pull video subtitles and process media

SkillExtended access228

README in short

The repository started as a Claude Code kit with Exa MCP, then gained a portable `skills/` folder for Codex and OpenClaw. The original Claude Code versions remain in `.claude/skills/`. Sample reports on markets and companies sit alongside.

README badge

Original SKILL.md

---
name: company-research
description: Company, competitor, and market intelligence research using Exa or an equivalent web-search/MCP tool. Use for requests to research a company, build competitor lists, find company news, summarize financial/funding/headcount/location facts, discover public LinkedIn/people profiles, or produce cited market landscape briefs in Codex, OpenClaw, and other SKILL.md-compatible agents.
---

# Company Research

## Core Rules

- Prefer Exa advanced web search when available. In Claude this is commonly `mcp__exa__web_search_advanced_exa`; in Codex/OpenClaw use the configured Exa MCP search tool or the host's web-search tool.
- Do not use weaker/basic Exa search tools when an advanced Exa search tool is available.
- Keep raw search noise out of the final answer. Work from distilled notes, merged entities, and citations.
- Use independent research passes only when the host supports them and the user/task policy allows them; otherwise run compact searches sequentially and summarize after each batch.
- Cite primary or high-quality sources for factual claims, especially current facts such as funding, revenue, executives, headcount, product launches, and pricing.

## Research Workflow

1. **Clarify scope if needed**
   - Identify target company, geography, sector, time horizon, and requested depth.
   - If the user gives an ambiguous quantity, ask for the desired number of companies or choose a reasonable default and state it.

2. **Plan query variants**
   - Generate 2-3 phrasings per topic to improve coverage.
   - Include local-language terms for non-English markets.
   - Mix source intents: official company pages, news, databases, social profiles, and analyst/industry sources.

3. **Run focused searches**
   - Use `company` category for homepages and metadata when Exa supports categories.
   - Use `news` for press coverage, launches, funding, earnings, partnerships, and recent events.
   - Use `people` for public LinkedIn-style profiles and leadership discovery.
   - Use `tweet` or social search only for brand/social presence when relevant.

4. **Merge and deduplicate**
   - Normalize company names, domains, subsidiaries, and brands.
   - Separate confirmed facts from inference.
   - Flag uncertain or conflicting data instead of silently choosing one source.

5. **Synthesize output**
   - Produce the format the user asked for: brief, table, competitor map, JSON, or full report.
   - Include citations next to key claims.
   - End with gaps, assumptions, and recommended follow-up searches when evidence is incomplete.

## Dynamic Search Depth

- "A few" / quick scan: 10-20 results total.
- Standard competitor/company brief: 20-50 results total.
- Comprehensive landscape: 50-100 results total across variants.
- User-specified number: match the requested number.

Report a problem