| name | exa |
| description | Web search, content crawling, code context lookup, company research, and deep AI research via Exa. Use when the user needs to search the web, fetch page content from a URL, find code examples or documentation, research a company, or run a deep research report on a complex topic. |
Exa
Web search, crawling, code context, and deep research via the Exa MCP server (exa-mcp-server).
Setup
- MCP server:
exa-mcp-server (npm, stdio transport)
- Config:
~/.pi/agent/mcp.json → mcpServers.exa
- Auth:
EXA_API_KEY in ~/.zshrc.local, interpolated via ${EXA_API_KEY}
- Mode: Proxy (accessed through the
mcp gateway tool)
When to Use
- Search the web for any topic
- Fetch full content from a known URL
- Find code examples, docs, and programming solutions
- Research a company
- Run deep AI research reports on complex topics
Available Tools
Access all tools through the mcp gateway:
mcp({ server: "exa" }) # List available tools
mcp({ search: "exa" }) # Search for exa tools
mcp({ describe: "tool_name" }) # Show tool details
web_search_exa (default, enabled)
Search the web for any topic. Returns clean text content from top results.
mcp({ tool: "web_search_exa", args: '{"query": "latest AI agent frameworks", "numResults": 5}' })
| Parameter | Required | Description |
|---|
query | yes | Search query |
numResults | no | Number of results (default: 8) |
livecrawl | no | fallback (default) or preferred |
type | no | auto (default) or fast |
contextMaxCharacters | no | Max characters for context (default: 10000) |
get_code_context_exa (default, enabled)
Find code examples, documentation, and programming solutions from GitHub, Stack Overflow, and docs.
mcp({ tool: "get_code_context_exa", args: '{"query": "React useState hook examples", "tokensNum": 10000}' })
| Parameter | Required | Description |
|---|
query | yes | Programming search query |
tokensNum | no | Tokens to return, 1000–50000 (default: 5000) |
crawling_exa (default, enabled)
Get the full content of a specific webpage from a known URL.
mcp({ tool: "crawling_exa", args: '{"url": "https://example.com/docs"}' })
| Parameter | Required | Description |
|---|
url | yes | URL to crawl |
maxCharacters | no | Max characters to extract (default: 3000) |
company_research_exa (optional)
Research a company for business information, news, and insights.
| Parameter | Required | Description |
|---|
companyName | yes | Company name |
numResults | no | Number of results (default: 3) |
deep_researcher_start / deep_researcher_check (optional)
Start an AI research agent for complex topics (15s–3min depending on model).
Start:
| Parameter | Required | Description |
|---|
instructions | yes | Detailed research question |
model | no | exa-research-fast (~15s), exa-research (15–45s), exa-research-pro (45s–3min) |
Check:
| Parameter | Required | Description |
|---|
researchId | yes | Research ID from start |
Keep calling check until status is completed.
Env Vars
| Variable | Location | Description |
|---|
EXA_API_KEY | ~/.zshrc.local | Exa API key from dashboard.exa.ai |
Troubleshooting
Server not connecting
- Verify
EXA_API_KEY is set: echo $EXA_API_KEY
- Run
source ~/.zshrc.local
- Restart Pi
- Check:
mcp({ server: "exa" })
Rate limit errors (429)
Tools not appearing
- Restart Pi after config changes
- Verify
~/.pi/agent/mcp.json is valid JSON
- Try
mcp({ connect: "exa" }) to force reconnect
