Docs

Steve MCP documentation

Steve MCP is a Model Context Protocol server that gives your AI assistant live competitor data: news, reviews, website changes, social activity and ads, with a source url and date on every item. Set it up below, then use the method reference for what your assistant can call.

( The server )

• Server namesteve-agent
• Endpointhttps://app.hiresteve.ai/api/agent/mcp
• TransportStreamable HTTP, stateless, JSON responses
• AuthAuthorization: Bearer <access token>
• ScopeOne person in one organization

( Setup, per assistant )

One pasted line in every client, your assistant sets everything up for you. No download, no call, no IT ticket. Pick yours:

( The methods )

Eighteen methods in eight groups. Everything is scoped to a project: call list_projects first, it maps the project name the user said to the id every other method needs. See all methods on one page, or jump to a group:

• Projects3 methodsA project is the container everything else lives in: an anchor company, its competitors, and the feeds collected for them. Every other method takes a projectId from list_projects.list_projects · prepare_create_project · create_project
• Competitors3 methodsCompetitors are the companies a project watches. Adding one starts the same background pipeline the app starts: reviews, news, company research, page discovery, social-handle discovery and starter battle cards.list_competitors · prepare_add_competitor · add_competitor
• Tracked accounts4 methodsThe social profiles a project scrapes. The account url stored here is the exact key the social read methods take; rebuilding it from a handle will not match.list_tracked_accounts · set_linkedin_handle · set_x_handle · set_instagram_handle
• News1 methodNews articles collected for a project's tracked competitors, newest first, each with its source url and publish date.get_news
• Reviews1 methodCustomer reviews collected for one competitor at a time, across G2, Capterra, Trustpilot and any custom sources, each review with its url and publish date.get_reviews
• Website changes1 methodDated changes observed on competitor websites: pricing pages, messaging, feature pages. Each change carries the page url it was seen on and when.get_website_changes
• Social posts3 methodsPosts scraped from tracked accounts and competitor domains, per platform. All three methods share one shape; scope by competitorId, by a tracked account url, or read everything.get_linkedin_posts · get_twitter_posts · get_instagram_posts
• Ads2 methodsAds collected from the Google Transparency Center and the Meta Ad Library. A competitor produces Google ads once its googleAdvertiserId is set, Meta ads once its facebookPageId is set; list_competitors shows both.get_google_ads · get_meta_ads

( Conventions )

• ›Receipts. Every item a read method returns carries its source url and date. No receipt, no claim; the provenance policy covers where the data comes from.
• ›Names, not ids. Responses carry ids for chaining calls, but they are never meant to be shown to the user.
• ›Pagination. List methods take limit and cursor and return nextCursor and hasMore. Out-of-range limits are clamped, not rejected.
• ›Next steps. Every successful response suggests what makes sense next as questions to ask the user, never calls to make unprompted.
• ›Honest emptiness. Empty results say why (nothing tracked, source not granted, nothing collected yet) instead of returning a bare empty list. See errors and limits.

( The free playbooks )

Four installable skills, free forever, no account. Each works on a bare assistant and honestly labels what it can't see. Browse them in the skills library.

( Trial and billing )

• ✓3 days free, everything included, ask something we couldn't have staged.
• ✓Then a flat monthly price: $49 Starter, $99 Pro, $199 Team. No annual invoice.
• ✓Cancel in one click.