# UnifAPI Agents: a folder of marketing skills for MCP clients, backed by a hosted data API > The unifapi-agent/agents repository is a set of SKILL.md folders that teach Claude, ChatGPT, Codex, OpenClaw, Hermes and other MCP clients how to run SEO audits, AI-visibility checks, local SEO research, influencer pricing, social listening and competitive intelligence. The skills are free and MIT licensed, but the data comes from UnifAPI's hosted public-data APIs over OAuth, so the useful part depends on a service you do not run yourself. **unifapi-agent/agents** — Open-source marketing agents for Claude, ChatGPT, Codex, OpenClaw & Hermes. One plugin: SEO audits, GEO / AI-visibility, local SEO, KOL pricing, social listening & competitive intelligence from read-only public data over MCP. - Repository: https://github.com/unifapi-agent/agents - Website: https://unifapi.com - Stars: 599 · Forks: 116 - Language: JavaScript - License: MIT - Published: 2026-10-22 · Updated: 2026-10-22 · Language: en - Canonical page: https://hysenlabs.com/projects/unifapi-agent-agents ## A prompt pack whose value depends on a service you do not run UnifAPI Agents is described as open-source marketing agents for Claude, ChatGPT, Codex, OpenClaw, Hermes, and any MCP client. The framing that matters is in the README itself: each agent is a folder of SKILL.md skills over UnifAPI's public-data APIs, and the whole thing is read-only, eyes not hands, so they research and price and never post. So the split is clear. The skills, the agent structure and the AGENTS.md map are the open source part, and they are MIT licensed. The data is the hosted part, and it comes from UnifAPI at https://mcp.unifapi.com over a remote streamable HTTP MCP server with OAuth. New workspaces get free trial credits. That is a genuinely different shape from most open source tooling in this space, and it is worth naming plainly before anything else. You are not adopting a library you install and run. You are adopting a set of instructions that your assistant follows while calling someone else's API on your behalf, with that API's data model and that API's coverage limits. The read-only constraint is the part that makes the arrangement reasonable. An agent that can research, pull ranking positions, read competitor mentions and price creators is genuinely useful, and one that can post to those platforms is a different risk class. The README puts it as a slogan, eyes not hands, and that is also a hard boundary on the skill set. The skills are generated and versioned, and the README says so: to suggest a change, open an issue or PR, with the CONTRIBUTING.md file as the route. So the prompt text is the product, and changes to it go through the repository rather than through local edits. The primary language is JavaScript, the default branch is main, the last push was on 2026-09-05, and the repository has one release tag, 0.2.2. ## Nine role agents, six verticals and six platform agents The agent catalogue is organised in three layers, and the README explains the difference: Role Agents own the skills, while Vertical and Platform Agents recombine them by industry and by source. The role agents are the core set. The SEO Agent tracks rankings, audits SERPs, and researches keywords, competitors and backlinks as live records inside your assistant. The AI Visibility Agent measures whether your brand and domain get cited in AI answers across ChatGPT and AI search, and where competitors win the answer. The Local SEO Agent audits local-pack rankings, maps listings and reviews presence for one location or hundreds. Then the commercial-facing ones. The Influencer Marketing Agent plans, matches, prices, confirms, tracks and reports creator campaigns from public engagement evidence. The Content Strategy Agent turns public demand, meaning keywords, Reddit threads, YouTube and news, into a content plan with evidence. The Social Listening Agent monitors public mentions across X, Reddit, TikTok and news and gives you a brief when something meaningfully changes. The Social Selling Agent spots buying signals and warm openings across X and LinkedIn, then prepares evidence-backed outreach. The Competitive Intelligence Agent tracks what competitors ship and say across X, LinkedIn, YouTube, Reddit and news and explains what changed. And the Lead and Company Research Agent builds B2B account and company briefs from public LinkedIn and news signals. The vertical agents are the same capabilities pointed at a sector: real estate, med spa and aesthetics clinics, dental practices, law firms and attorneys, restaurants and hospitality, and home services and contractors. Each is described in terms of local rankings, treatment-content or practice-area content, reviews, social buzz and AI visibility, which is the same skill set with a different subject matter. The platform agents are the source dimension: X, LinkedIn, Reddit, TikTok, YouTube and Instagram, each described as read-only research inside your assistant. That three-layer structure is the design idea worth taking away, because it is what makes the catalogue extensible without rewriting prompts. A new agent is usually a recombination rather than new code. The full agent to skill to API map is in AGENTS.md in the repository root, and that is the file to read before adopting anything, because it is where the boundary between what a skill asks for and what the platform actually returns becomes visible. ## Installing means adding a remote MCP server, and OAuth is the only credential The install section describes the connection as OAuth with no API key to paste, and then gives a different recipe per client because the clients genuinely differ in what they can do. For clients that manage their own MCP servers, which the README lists as Claude Code, Codex, Cursor, OpenClaw and Goose, you paste one prompt into the assistant and it adds the server and runs the OAuth sign-in for you. The prompt is worth reading as a whole, because it states the three facts the assistant needs: the server is remote and speaks streamable HTTP, the URL is https://mcp.unifapi.com, and UnifAPI uses OAuth with a browser sign-in. Then the skills themselves are added separately with npx skills add unifapi-agent/agents. The web connector route covers Claude, ChatGPT, Perplexity and Grok with no terminal at all: you add a custom connector pointed at the same URL and authorise over OAuth. ChatGPT needs Developer Mode, reached through Settings, then Connectors, then Advanced, and you create the connector with the MCP server URL and OAuth. Codex ships a plugin manifest at .codex-plugin/plugin.json that bundles both the skills and the OAuth MCP server. If you would rather wire it by hand you add the server to your Codex config and log in through the CLI. ```toml [mcp_servers.unifapi] url = "https://mcp.unifapi.com" ``` Claude-compatible plugin hosts, which the README names as Claude Code, Claude Desktop and OpenClaw, can install the skills and the server together through the plugin marketplace. ```bash /plugin marketplace add unifapi-agent/agents /plugin install unifapi@unifapi # that's plugin@marketplace ``` And any other MCP client just points at the URL with OAuth and adds the skills with the same npx command. Step-by-step pages for every supported client live in the connect/ directory in the repository. ## The audit prompt is a specification of what the output should contain After connecting, the README gives the example prompt to send, and it is short but dense with requirements. Three deliverables are named: our live organic positions, the competing pages, and a prioritized fix list. Two keywords are given, which is how the audit is scoped rather than being a generic site review. And there is a hard requirement attached to every claim, cite every finding. That last clause is the important one for anyone adopting this for real work. A marketing audit whose findings are unciteable is marketing fluff, and the prompt encodes the expectation that the assistant attributes each claim to a source it retrieved rather than to its own reasoning. The README's opening phrase, research and price, never post, is the same idea applied to the whole catalogue. The example also shows the shape of a skill at work. It names a domain, a set of query terms, the data to pull and the format of the output, and it leaves the actual API calls to the skill definition in the SKILL.md folder. The agent is the model following those instructions, and the UnifAPI server is the thing that answers. The installation is also self-describing in a useful way. You do not paste an API key, so there is no secret to rotate in your own configuration and no key to leak from your shell history or your agent's context window. The trade is that authorisation is granted in the browser to UnifAPI, and the scope of that grant is UnifAPI's and the client's, not yours to configure in a config file. ## What the read-only boundary does and does not protect you from The README makes the read-only claim twice, once for the overall set and once per platform agent, and it is worth taking the claim seriously while also seeing where it stops. What it means concretely is that no skill in the catalogue can post to X, publish to LinkedIn, comment on Reddit, upload to TikTok, or send anything through the Instagram graph. The platform agents are described as research of public profiles, posts, communities, trends, jobs, threads, creators, videos, hashtags and comments. The Social Selling Agent prepares outreach; it does not send it. The Influencer Marketing Agent confirms and tracks creator campaigns; it does not negotiate them. That is a meaningful boundary, and it is the reason an assistant with marketing reach is being handed a plugin whose whole purpose is to study marketing. The tools that would let an agent act on a social account are not in the box. What it does not cover is the data leaving your machine. The queries you ask about, the domains you audit, the competitor names you research and the creators you price are all arguments passed to UnifAPI's server, and the README says the agents work from live public data. So the read-only property limits what the agent does with the data it retrieves; it says nothing about who sees the questions you asked. That is not a criticism of the design, it is a description of it. If your marketing research involves a confidential product launch, an unannounced partnership, a client name or a market you are not planning to discuss publicly, that context is being sent to a third-party API as part of the query, and you should factor that into what you ask. The skill catalogue also names the data sources it covers, which is useful for scoping. X, Reddit, TikTok and news for listening, X and LinkedIn for social selling signals, YouTube and Reddit and news for competitive intelligence, LinkedIn and news for lead research, and search results and AI answers for the SEO and visibility agents. ## Where UnifAPI Agents is the wrong tool The first wrong-tool case is a team that needs its own data or its own infrastructure. Every skill resolves to a UnifAPI API, the MCP server is remote, and the OAuth flow is the only credential path offered. If your ranking data comes from your own Search Console or your own analytics, or if you need to run this inside a network boundary, there is no offline mode and no local data source documented. The MIT licence covers the skill text, not the data. The second case is a team that cannot accept a hosted dependency in an agent loop. An assistant making repeated calls to a remote API on every question has latency, cost and availability considerations that a local script does not, and the README says nothing about caching, rate limits or offline behaviour beyond the trial credits. The third case is a role the catalogue does not cover. The verticals are real estate, med spa, dental, law, restaurant and home services, and the roles are SEO, AI visibility, local SEO, influencer marketing, content strategy, social listening, social selling, competitive intelligence and lead research. If your marketing motion is email deliverability, paid acquisition, product analytics or lifecycle campaigns, there is no skill for it, and the architecture does not obviously extend to those without the underlying APIs existing. The fourth case is a job that requires judgement rather than retrieval. The competitive intelligence agent tracks what competitors ship and say and explains what changed, and the social selling agent spots buying signals and prepares openings. Those are retrieval plus assembly. The strategic call of whether to pursue a market is not something this catalogue makes, and the read-only framing means it does not try. On maintenance and cost, the practical questions are about the service rather than the repository. The skills are generated and versioned, so an upstream API change is absorbed by regenerating them, and the repository is the place to raise that. But the data access itself is a subscription, the README mentions free trial credits for new workspaces and nothing about pricing, and the README does not document a rate limit, a caching layer or what happens when a quota is exhausted. A team building a repeatable audit process on top of this should settle those questions before committing to it, because the dependency is the API and not the prompts. The licence is MIT with a LICENSE file at the root, so the skill files themselves are yours to fork and adapt. What you cannot fork is the API, and that is the part that decides whether the whole arrangement works for you. ## Conclusion UnifAPI Agents suits a marketer or engineer who wants an assistant to run a structured SEO or AI-visibility audit without building the data pipeline, because the skill folders are readable Markdown and the phase list in the install prompt shows the shape of the output. It does not suit a team that needs its own data or a self-hosted deployment, since every skill calls UnifAPI's hosted MCP server at https://mcp.unifapi.com and the OAuth flow is the only credential mechanism offered. Before adopting, check what UnifAPI charges after the trial credits and read AGENTS.md, which maps each agent to its skills and underlying APIs, because that mapping is what tells you whether a skill is worth the dependency. ## FAQ ### What is UnifAPI Agents? It is a collection of open source marketing agents, each a folder of SKILL.md skills, for Claude, ChatGPT, Codex, OpenClaw, Hermes and any MCP client. The skills cover SEO audits, AI-visibility checks, local SEO, influencer pricing, social listening and competitive intelligence, working from live public data. ### How do I install the UnifAPI Agents skills? For clients that manage their own MCP servers you paste a prompt that adds the remote server at https://mcp.unifapi.com and runs the OAuth sign-in, then add the skills with npx skills add unifapi-agent/agents. Claude-compatible plugin hosts can install both at once with /plugin marketplace add unifapi-agent/agents followed by /plugin install unifapi@unifapi. ### Do the UnifAPI agents post anything on my behalf? No. The README describes them as read-only, eyes not hands, so they research and price and never post. The social selling agent prepares evidence-backed outreach and the influencer agent confirms and tracks campaigns, but nothing is published to a social account. ### What agents does the UnifAPI Agents repository include? Nine role agents covering SEO, AI visibility, local SEO, influencer marketing, content strategy, social listening, social selling, competitive intelligence and lead and company research, six vertical agents for real estate, med spa, dental, law, restaurant and home services, and six platform agents for X, LinkedIn, Reddit, TikTok, YouTube and Instagram. ### Does UnifAPI Agents need an API key? No. The README states that UnifAPI uses OAuth with a browser sign-in and that there is no API key to paste, because it is read-only public data. New workspaces get free trial credits. ### What license is the UnifAPI Agents repository released under? The skills are released under the MIT license, with the LICENSE file at the root of the repository. The hosted API the skills call is a separate commercial service, and the README does not document its pricing. ## Sources - [License: MIT](https://github.com/unifapi-agent/agents/blob/main/LICENSE) - [Project website](https://unifapi.com) - [README](https://github.com/unifapi-agent/agents/blob/main/README.md) - [Releases](https://github.com/unifapi-agent/agents/releases) - [unifapi-agent/agents on GitHub](https://github.com/unifapi-agent/agents) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/unifapi-agent-agents