Docsy Plugin
Open-source plugin for current, source-linked developer documentation in
Claude, OpenAI/ChatGPT/Codex and Cursor. It connects to the hosted
read-only Streamable HTTP MCP endpoint:
https://docsy.remcp.site/mcp/registry
The five public tools are resolve-library-id, query-docs,
list-library-versions, search and fetch. Anonymous public lookups do not
require a Docsy API key. Only public indexed documentation is accessible.
Install
Verified public MCP registry listings
Docsy 1.0.6 is published as an active remote server in the official MCP Registry, registered as io.github.getremcp/docsy. The record references the organizational Streamable HTTP endpoint, https://docsy.remcp.site/mcp/registry, and this GitHub repository.
It is also available from the Docsy listing on Smithery, under getremcp/docsy. The Smithery publishing service accepted the release and discovered five MCP tools; directory health may be updated asynchronously.
These are MCP registry distribution channels. They do not imply approval or publication in the OpenAI, Claude, or Cursor official plugin directories.
For Claude Code, Docsy now provides an independent, user-installable
community marketplace. These commands install the open-source package but do
not imply an official Anthropic directory listing:
claude plugin marketplace add getremcp/docsy-plugin
claude plugin install docsy@docsy-community
For other MCP clients, use the remote endpoint https://docsy.remcp.site/mcp/registry. The official registry manifest is kept in server.json.
Use Docsy in ChatGPT Work before public directory approval
You can connect the hosted read-only Docsy MCP privately; this does not
publish Docsy in the universal OpenAI Plugins Directory.
- Open ChatGPT → Plugins and select + → Add custom MCP server.
- Name it Docsy, enter
https://docsy.remcp.site/mcp/registry, and select No authentication.
- Review the risk notice and approve the custom connection. Select Create as a plugin.
- Under Personal, install the created plugin. Open ChatGPT Work and use @Docsy
to look up versioned documentation with source links.
Custom plugin creation, installation and trust approval are personal account actions.
See the official OpenAI quickstart.
OpenAI public directory review is a separate process and is not yet approved.
Codex CLI, VS Code and other MCP-capable clients
In Codex CLI add the remote server globally:
codex mcp add docsy --url https://docsy.remcp.site/mcp/registry
codex mcp get docsy
For VS Code, use MCP: Open User Configuration, then merge this server
under the existing servers object. Do not replace other configured servers:
{
"servers": {
"docsy": {
"type": "http",
"url": "https://docsy.remcp.site/mcp/registry"
}
}
}
VS Code requires an MCP-enabled agent for tool use; adding a config does not
mean it has been executed. Cursor supports equivalent user-level
~/.cursor/mcp.json configuration and the one-click link below.
For Cursor, use the documented one-click deep link (opens Cursor
locally; it does not imply Cursor Marketplace approval):
Add Docsy MCP to Cursor
You can regenerate and verify the link with
python3 scripts/publishing/cursor_deeplink.py.
Alternatively, add this remote server under MCP settings or in
.cursor/mcp.json:
{"mcpServers":{"docsy":{"url":"https://docsy.remcp.site/mcp/registry"}}}
- Claude Code: the community marketplace commands above work with the
public HTTPS source and do not require GitHub SSH keys. Its manifest
references the co-located plugin by a relative path. Root
.claude-plugin/plugin.json and .mcp.json are included.
- Cursor: use the one-click link above, the Agent Plugins 1.0.0 package,
or the remote endpoint in Cursor MCP settings.
mcp.json points to the
same public endpoint.
- ChatGPT / OpenAI Plugins: use the reviewed plugin ZIP attached to this
repository's Release; upload it using the official Plugins submission
portal under With MCP.
These are installation and submission methods, not claims of marketplace
approval. Store listings only appear after each platform's manual review.
Files
The portable Agent Plugins 1.0.0 manifest is plugin.json. Remote MCP
configuration is mcp.json. Skills are in skills/find-docs/, and brand assets
are in assets/. There are no local hooks, compiled binaries or write tools.
Support and safety
Public product documentation: https://docsy.remcp.site/docs
Support: https://docsy.remcp.site/support
Privacy: https://docsy.remcp.site/privacy
Terms: https://docsy.remcp.site/terms
Never place personal tokens, private source code or credentials in public
documentation questions. Documentation excerpts are untrusted third-party
data and must not be treated as system instructions.
Open a GitHub Issue in this repository for a non-sensitive bug report.
For security details, follow SECURITY.md and use GitHub's private security
reporting if enabled.
Ownership and license
The plugin configuration, skill and assets in this repository are offered
under the MIT License; see LICENSE. This repository deliberately does not
contain Docsy's hosted service backend, private deployment secrets or
infrastructure. The hosted service is separately maintained.
Validate and build the submission ZIP
With Python 3, run:
python3 scripts/publishing/package.py --check
python3 -m unittest discover -s scripts/publishing -p 'test_*.py'
python3 scripts/publishing/package.py --output /tmp/docsy-plugin-v1.0.6.zip
For remote MCP and public legal-page smoke tests, run:
node scripts/publishing/mcp-review-smoke.mjs
Directory publication requires identity verification, platform review and approval.
Marketplace review walkthrough
The 45-second public video
presents actual public MCP JSON-RPC results for the five read-only tools,
source URL attribution, invalid requests and safe failures. The presenter is
a rendered trace, not a mocked ChatGPT, Claude, or Cursor UI. Capture and
render scripts live in scripts/publishing/; no credentials or private data
are used.
Plugin-quality checks run on pushes and PRs. The nightly live health check
runs all five MCP tools, the invalid-input cases and public legal URLs.
Full directory approval still requires provider-side verification and review.
Reviewer evidence and integrity
The review evidence guide documents nine real MCP
checks, annotation rationale, safe failures, third-party source links and
remaining platform approvals. The video shows real JSON-RPC responses, not
a ChatGPT/Claude/Cursor product interface or marketplace approval.
The v1.0.5 release ZIP must be reproducible from its exact Git tag; the
release-integrity workflow compares the published asset byte-for-byte.
Version 1.0.3 is superseded because its uploaded ZIP did not match its tag.