MCP field guide / 01
Connect your AI client to live startup jobs.
Search current openings, investigate startups, and verify the source—without scraping pages or handing an agent write access.
Production endpoint
https://jobswithstartups.com/mcpNo API key is required. Your client must support modern MCP Streamable HTTP or use the bridge shown below.
Start here
Quick start
Connection takes one configuration block and a fresh client session. Choose the format your client supports.
- 01
Choose a connection
Use the URL directly in clients with remote MCP support. Use the stdio bridge for clients that expect a local command.
- 02
Add the server and restart
Paste the matching configuration into your client, save it, then start a new session so the four tools are discovered.
- 03
Ask for a sourced result
Tell your client to use Jobs with Startups and include canonical job or company URLs in its answer.
Direct remote connection
{
"mcpServers": {
"jobs-with-startups": {
"url": "https://jobswithstartups.com/mcp"
}
}
}Local stdio bridge
{
"mcpServers": {
"jobs-with-startups": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://jobswithstartups.com/mcp"
]
}
}
}The interface
Tool reference
Start broad with search, then use a detail tool to verify the record your agent is discussing.
01
search_jobs
Discover
Find active roles by keywords, workplace type, employment type, industry, company size, investor, funding stage, or Y Combinator status.
Returns · Job summaries, application links, and canonical URLs
02
get_job
Verify
Open one job by its canonical slug when you need the description, compensation, source link, and startup context.
Returns · A bounded, plain-text job record
03
search_startups
Research
Search active startups by keywords, investor, or funding stage before comparing companies or looking for their openings.
Returns · Startup summaries and matching counts
04
get_startup
Investigate
Inspect one public startup profile, including active jobs and selected funding, investor, founder, news, and culture research.
Returns · A compact company dossier with current jobs
Put it to work
Prompt recipes
Good prompts state the decision you are making, the filters that matter, and the source evidence you want returned.
01
Build a focused job shortlist
“Find up to 10 remote product design roles at seed or Series A startups. Group them by company and include the source application link.”
02
Research before you apply
“Research the startups behind these job results. Compare their funding, work policy, recent news, and active role count. Clearly mark unavailable fields.”
03
Verify a saved opportunity
“Open this Jobs with Startups job URL, verify that the listing is still active, and summarize the responsibilities, compensation, and application destination.”
Trust the boundary
What the server will—and won’t—do
The MCP is intentionally narrower than the website’s internal systems. That keeps agent access useful, predictable, and safe.
Reads current public job and startup data; it cannot add, edit, sync, or apply to jobs.
Searches return at most 20 records per call and long descriptions are deliberately bounded.
Usage analytics record only the tool, outcome, coarse duration, and coarse result count.
Applications continue on the employer’s job board; always verify final requirements at the source.
If something looks off
Troubleshooting
Most connection issues come from client configuration or an old session rather than the server itself.
The server does not appear
Confirm the endpoint ends in /mcp, save the configuration, and fully restart or open a new client session.
My client only accepts commands
Use the mcp-remote configuration above. It translates the remote Streamable HTTP endpoint into a local stdio process.
A search returned no results
Remove one filter at a time or broaden the keywords. An empty result is valid and does not mean the connection failed.
A tool asks me to retry
Retry once. The server returns a safe generic error when its public data source is temporarily unavailable.
Give your agent a better source.
Explore the directory first, then connect the same current data to the AI client where you already work.