Snagajob

Name: Snagajob MCP Server
Availability: InStock
Author: Snagajob

HIPAA CompliantSOC2 ReadyISO 27001 Ready

Productivity2 toolsNo Auth Required

Search US hourly job listings from an AI agent with no account and no sign-in. Two read-only tools, nine filters on search_jobs, and an endpoint on jobget.com because JobGet owns Snagajob. Applying still happens on Snagajob's site.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on 8 September 2026.

Anthropic states this reflects the level of review a connector received, not a security audit.

Connect Snagajob via MCP

https://platform.jobget.com/jobs-mcp/mcp

Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.

Use in Agentman

Connect Snagajob once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.

Open in Agentman Studio

Snagajob Tools & Capabilities (2)

search_jobsRead-only

Search for hourly and local job postings near a location. Returns a ranked list of matching jobs plus facet breakdowns (by industry, company, pay band, employment type) you can use to refine. Use this when the user asks about jobs, hiring, or finding work in a specific area.

get_job_detailsRead-only

Fetch the full details for a single job posting by ID. Use this when the user wants to know more about a specific job from a search result — full description, requirements, benefits, schedule, apply link.

Read from the server on 2026-09-08, including each tool's own safety annotations.

Limits

Applying is not in scope, by construction. Both tools are annotated read-only, neither accepts application data, and the widget's declared CSP permits one redirect target. To apply you leave the conversation for www.snagajob.com, where an account may be required.
location is mandatory. search_jobs will not run without it, so there is no "show me anything anywhere" query. That is a privacy benefit and a usability cost in the same field.
Twenty-five results per call, fifty miles maximum radius. Both ceilings are in the live schema. A wider or deeper survey needs several calls, and any claim your agent makes about "the whole local market" rests on that sample plus the facet counts.
US hourly work only, as far as we can tell. Every source describes a US hourly marketplace, and the schema has no country parameter — location_resolved returns only city, state and zip. Snagajob does not state a country restriction in the documentation we read, so this is our reading of the schema, not a vendor claim.
The declared widget domain does not resolve. Both tools name https://mcp.jobget.com as openai/widgetDomain; that hostname returned NXDOMAIN from 8.8.8.8 on 2026-09-08 while platform.jobget.com resolved normally. We could not determine the consequence.
The connector's own metadata is partly addressed to another client. Both tools carry openai/-namespaced fields alongside the vendor-neutral ui block. This is one server shipping to several clients, which is what MCP's _meta extension point is for, but it means some of what the server declares is not aimed at the reader's client.
We could not read the privacy policy. legal.snagajob.com renders its policy sections client-side and served no policy text to our fetch on 2026-09-08, so we cannot report what it says about technical data, IP addresses or AI assistants.
The vendor site blocks plain HTTP clients. www.snagajob.com and legal.snagajob.com returned HTTP 403 Cloudflare challenges to curl with a browser user-agent, including for the robots-declared integrations-sitemap.xml, while a browser-class fetcher retrieved the same pages at 200. The robots.txt file itself returns 200 to curl and allows every path we read, so this is bot-blocking rather than a policy signal.
No rate limit is published, and we did not stress the endpoint to find one. Absence of a documented limit is not evidence there is none.

Frequently asked questions

No. The endpoint answered an anonymous handshake with a full tool list and no credential of any kind, and Anthropic's directory records the server as authless. There is no OAuth flow, no token and no scope to grant. You only need a Snagajob account later, on Snagajob's own site, if the job you pick requires one to apply.

Because JobGet owns Snagajob. JobGet announced the acquisition on 15 November 2024 and said Snagajob would operate as a standalone entity within JobGet. Anthropic's directory names JobGet as the connector's author, and Snagajob's own site footers read Snagajob by JobGet. The host difference is corporate structure, not a mismatched or third-party endpoint.

No. Both tools are annotated read-only by the server, and neither takes a resume, a name or any application field. Snagajob's own integration page says Claude returns direct Apply on Snagajob links, and the connector's widget declares snagajob.com as its only permitted redirect target. Applying is a link you follow in your browser.

Nine parameters, read from the live tool schema. Location is the only required one. The rest are a free-text query, a radius up to fifty miles, employment type, minimum hourly pay, up to twenty industries, shift, how recently the job was posted, and a result limit capped at twenty-five per call.

Twenty-five at most. The live schema caps the limit parameter at twenty-five, and the response also carries a total count plus facet breakdowns by industry, company, pay band and employment type. So an agent sees a sample and a summary of the wider result set, not every matching listing.

The tools do, and Snagajob publishes a ChatGPT setup page alongside its Claude one. But Anthropic's directory lists only Claude surfaces under works with, and both tools ship an interactive card widget whose declared host does not resolve in public DNS. A client without MCP Apps support receives the structured results as text instead.

Snagajob publishes no price, plan gate or quota for it, and with no sign-in there is no account to attach billing to. That is not a guarantee of no limits. Snagajob documents no rate limit anywhere we could find, and absence of a published limit is not the same as absence of a limit on the endpoint.