Actively
Read per-account sales intelligence — agent memory, decisions, signals — from an AI agent.
Run a marketing agency's social, SEO, ads, WordPress and email from an AI agent.
Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.
Paste it into any MCP client. Setup docs
Abency is a Spanish multi-brand marketing platform for agencies, and its MCP server connects that platform to Claude, ChatGPT and any MCP client. Anthropic's directory lists 47 tools that schedule social posts, run keyword research, read Search Console and analyse Google and Meta Ads across every brand an agency manages. Sign-in is OAuth or a bearer token.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
tools/list response was read, no parameter schemas are published here, and no safety annotations were observed. Both counts are second-hand.mcp, covers reads, writes, the delete and anything added later. The only narrowing mechanism Abency documents is a single-brand token, which is a token-issuance choice rather than a consent-time one.abency-mcp package on npm is marked deprecated in Catalan, directing users to the remote server and to a setup document at github.com/gginerb/social-media-app-new. That repository returned 404 on 2026-08-23, so the migration note's own link is broken. Use the remote endpoint above; the npm package is superseded.Yes, to whichever accounts a brand has connected. Abency documents create_post as create and schedule a post across Facebook, Instagram, LinkedIn, TikTok and YouTube. Its FAQ states a post is scheduled if you give a date and time and otherwise stays a draft until you confirm. So publishing needs a second instruction, but nothing external gates it.
Anthropic lists 47 and Abency documents 63, so neither number is the whole surface. The two lists disagree in both directions: Abency describes 22 tools Anthropic never lists, mostly WordPress, Brevo email and Tag Manager, while Anthropic lists 6 work-order and timesheet tools Abency's page never mentions. Expect closer to 69 distinct names.
One scope named mcp, which bounds nothing. Abency's RFC 9728 resource descriptor and its authorization server both advertise scopes_supported of exactly that single value, checked live on 2026-08-23. It names no object and no verb, so granting it grants whatever the server implements, including deleting a scheduled post and rewriting a WordPress page.
Not the real send, according to Abency. Its documentation describes brevo_create_campaign and brevo_update_campaign as working on drafts, and says of brevo_send_test that Claude never triggers the real send. But brevo_send_test does dispatch mail to addresses the agent names, and brevo_update_campaign edits recipients, so mail leaves the system.
The keyword research tools do. Abency's pricing page meters SEO by plan: the Básico plan includes 25 tracked keywords and 5 searches a month, Growth 100 and 20, Pro 250 and 50, plus monthly caps on AI keyword clusters and competitor analyses. No tool in either list reads a balance, so an agent cannot check remaining quota before spending it.
Yes, one tool deletes. delete_post appears in both Anthropic's listing and Abency's documentation, which describes it as cancelling a scheduled post. That is the only delete verb in either list. It is recoverable only by recreating the post, and the single mcp scope offers no way to withhold it while granting reads.
Yes, despite the documentation URL. Abency's nominated docs page is titled for Claude Code, but its connection page states Abency is published as an official connector in both the Claude and the ChatGPT directories and that all three clients get the same tools. Anthropic's entry lists Claude, Claude API, Claude Code and Claude Desktop.
Abency says no, and that claim is the vendor's rather than ours. Its documentation states each token is bound to one organization and that the backend validates on every call that the requested brand belongs to it. We could not test this, because testing would require credentials. Treat the isolation boundary as documented, not as verified.
Yes, an SEO module is required, though Abency names two different starting plans. Its Claude Code page says the GSC, keyword research and tracking tools need a plan with an SEO module, giving Básico or higher in Spanish and Starter or higher in English. Brands, connections, posts and ads tools work on any plan.
https://app.abency.com/mcp returned HTTP 401 with WWW-Authenticate: Bearer resource_metadata=...; RFC 9728 metadata at https://app.abency.com/.well-known/oauth-protected-resource and authorization server metadata at https://app.abency.com/.well-known/oauth-authorization-server (2026-08-23). A control path under the same prefix returned 404, so these are genuine documents and not a catch-all. · retrieved 2026-08-23abency-mcp 1.2.0 (retrieved 2026-08-23) · retrieved 2026-08-23robots.txt, which disallows only /admin, /dashboard and /organization and carries no Content-Signal (retrieved 2026-08-23) · retrieved 2026-08-23https://abency.com/privacy-policy, which is a 301 to this Spanish canonical URL. · retrieved 2026-08-23Connect Abency once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.