Skip to content

DocPin MCP

The report an agent builds has always been the hard part to hand over. The HTML exists, and then it sits in a chat window the next person cannot open, or in a file nobody can find three weeks later. DocPin is the narrow product on the other side of that gap: one tool call publishes the page to a link that stays the same across every later version, and a second set of tools decides who is allowed to open it.

Server type: Remote (Streamable HTTP)
Endpoint:    https://mcp.docpin.app/mcp
Auth:        OAuth 2.0, dynamic client registration, PKCE
Scopes:      publish, settings
Registry:    app.docpin/docpin 1.0.0 (active, published 2026-11-07T17:47:59Z)
REST:        https://api.docpin.app with an API key
Link forms:  acme.docpin.it/{slug} · docs.acme.com/{slug} (Pro) · docpin.it/Ab3dE7
Plans:       Free (3 pages) · Pro $8 a month

What the endpoint says about itself

A keyless initialize on https://mcp.docpin.app/mcp answers 401 with

WWW-Authenticate: Bearer realm="OAuth",
  resource_metadata="https://mcp.docpin.app/.well-known/oauth-protected-resource/mcp",
  scope="publish settings"

and the document it names answers 200 with resource set to the MCP URL, the authorization server on the same host, scopes_supported of publish and settings, bearer_methods_supported of header, and resource_name of DocPin. That is the whole auth model in two reads: there is no API key to paste into a chat client, and the two scopes are the split that matters operationally. A connection approved without the settings tick can publish pages and cannot touch branding, domains, members or billing.

The registry record is a single active version, 1.0.0, published four minutes before the submission was filed, with one streamable-http remote and the website docpin.app.

The tools, by the job they do

Publishing. publish_page takes html, or files as [{path, content}] with an index.html inside, and returns the links. Republishing the same slug publishes a new version behind the same link, which is the single behaviour that makes this usable for a recurring report: the URL in a partner's bookmarks does not rot. A draft returns url: null and a previewUrl that works for about an hour and is not for sharing. For large pages there is create_upload_link, a one-time link valid 15 minutes that takes a file up to 25 MB (a .zip with an index.html works) and prints the links as JSON as soon as the upload lands, which is the route for anything an agent can write to disk faster than it can inflate a JSON string.

Who can open it. set_access takes team, public, password, people (emails and @domains) or inherit. Password pages and people-only pages need Pro, and the inherit value is what makes boards coherent: pages in a board follow the board. set_draft with draft: false is the share step and it requires an access value, so the agent has to ask the person who should be allowed in rather than defaulting to public. rename_page keeps the short link and redirects the old addresses, and move_page puts a page in or out of a board.

What happened after. list_versions and restore_version give history and a rollback. get_page_views reports how often a page was opened and, on a people-only page, by whom. That last pair is the difference between publishing and sending: a report nobody opened is a fact you can see.

Kept current without republishing. Every page carries a JSON store, up to 1 MB, read in the browser with docpin.data() or docpin.onData(cb, { every }). set_page_data merges by default (a null removes a key), replace: true replaces the object, and get_page_data first is the documented way to avoid losing keys. This is what a dashboard actually needs: the layout and the numbers are separate artifacts, so a nightly job updates the figures and the page keeps its link, its version history and its access rules untouched. Writes are capped at about 30 a minute per workspace (429 over that) and the store is 413 past 1 MB.

Workspace settings. get_workspace_settings returns the plan, the new-page defaults, branding, custom domains with DNS status, members and invites, and billing in one call. add_custom_domain returns the CNAME to set plus a one-click link where the DNS provider supports it, check_custom_domain is open to any member, and set_branding_image takes a URL or base64. get_billing_link returns a Stripe link with interval of month or year: the person pays on Stripe themselves. Deleting a workspace, creating API keys and delete_page stay behind an explicit human request, which is the right side of the line for an agent to be on.

The published page, and what it cannot do

The page runs in a sandboxed frame on its own address under docpinusercontent.com, one address per published version, with the surrounding DocPin page carrying the link, the badge and the password or sign-in screen. localStorage and sessionStorage work, kept per page and per visitor, and survive both reloads and new versions. Cookies and IndexedDB are blocked, service workers with them, and the ?query and #hash sync with the address bar so state written with history.replaceState is shareable. The vendor's own checklist for AI-generated pages is the practical version of all of this: one self-contained HTML document, handle empty data, wrap localStorage in try/catch, and avoid anything needing the camera or location.

That means a DocPin page is a place to read a report, filter a table and keep a view in the URL. It is not a host for an authenticated web app, and it is not a file store. Page data is as visible as the page it feeds, which the vendor states plainly: never put secrets in it.

Why it earns a guide rather than a hold

The artifact shelf already has owners for two adjacent jobs. HTML Pub publishes AI-generated HTML to a domain through the Leadpages ecosystem, and Shareglow fixes how a link renders when it is pasted somewhere. Neither of them answers the question an operator asks the second time: who is allowed to open this, and did they. DocPin's access modes, version history with restore, view log and the data store that updates a dashboard without a republish are four capabilities in one product, and the stable link across republishes is the mechanical part that makes the other three usable for anything recurring.

For any team whose deliverable to a client is a report, the pattern is: the agent builds the page, publish_page on a named slug, set_access to the client's domain, get_page_views as the follow-up signal, and set_page_data from whatever job recomputes the numbers. That is a delivery loop, not a hosting account.

Limits Recorded

  • Free is 3 pages with workspace links, public pages, secret links and page data. Pro at $8 a month adds unlimited pages, custom domains, password and people-only pages, and a smaller badge.
  • Page data is 1 MB per page and about 30 writes a minute per workspace; both limits return their own status rather than failing quietly.
  • Upload links expire in 15 minutes; draft preview links last about an hour and are explicitly not for sharing.
  • The MCP endpoint is OAuth-only. An agent that cannot complete a browser sign-in cannot use it, whatever its key situation.
  • Page data, view counts, version links and access lists all live with the vendor. This is a hosting product, not a connector to data you already own.
  • Hosted and priced as a service with no public repository, so the sandbox behaviour is documented rather than auditable. Recorded as a limit, not a criticism.
  • Read on 2026-11-07. Re-read https://docpin.app/docs/index.md before quoting the plans, which changed once already.

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

See Also

  • HTML Pub MCP - publish an agent's HTML to your own domain through Leadpages
  • Shareglow MCP - Open Graph previews, branded images and click stats for the links you publish
  • SumOffice MCP - the other half of the artifact loop: edit a workbook and return it as a download
  • ReportGem MCP - the report itself rather than the link it lives on

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing