slashdocs CLI
A CLI and skill for an agent to publish reports and specs as HTML documents that a team comments on, with every version kept
- 410 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The agent uploads report contents to a slashdocs server, and default rights on new documents are set by an administrator and can include view or edit access
- The CLI keeps sign-in credentials, and service account keys give access to documents with no human involved
Install
- Works in
Manual install
git clone https://github.com/razorpay/slashdocs.git && cd slashdocs && make buildBuilding from source produces bin/slashdocs-server and bin/slashdocs. Ready CLI archives sit in the repository releases, and the server serves an installer at /install.sh.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent uploads report contents to a slashdocs server, and default rights on new documents are set by an administrator and can include view or edit access
- The CLI keeps sign-in credentials, and service account keys give access to documents with no human involved
This is third-party code. Review the repository files before installing.
What it does
Razorpay's slashdocs publishes an agent's report as one self-contained HTML file instead of a wall of text in chat. People open the document in a browser, comment on exact words and share the link, while the agent reads the comments and publishes the next version. Every version is immutable, and concurrent edits are handled with optimistic concurrency. The agent works through the `slashdocs` command line. `docs create` uploads a file, `docs pull` and `docs push` fetch and update a document, `docs comments` lists, creates and resolves comments, and `docs list` searches the text of the latest version with `from:`, `in:`, `label:` and date filters. The `--json` flag gives structured output. Before creating a document the skill tells the agent to run `slashdocs guidelines`, which holds the rules for a single-file HTML, images and theme. There are three ways to connect. The `--local` mode runs on your computer with no server or login, but without sharing, version history, search or notifications. The full mode needs a slashdocs server: one Go binary with an embedded React interface, Postgres and S3-compatible storage. For agents without the CLI the server exposes MCP at `/mcp` with browser sign-in.
Who it is for. For teams where agents write long artifacts such as architecture reviews, incident write-ups and specs, and people need to discuss them and send edits back.
Good fit when
- An agent writes a report, incident write-up or spec that is awkward to read in chat
- Colleagues need to comment on exact words of a document and resolve comments, and the agent then reads them and revises
- You need version history for an agent's document and the ability to return to an earlier one
- You want to keep browser test results, dashboards and Mermaid or SVG diagrams as portable HTML files
- You want to try it with no server or login: `--local` mode keeps documents in the `~/slashdocs` folder
Not a fit when
- The user asks for Markdown, plain text, a PDF or an answer in chat: in those cases the skill does not publish to slashdocs
- You need real collaboration without your own server: sharing, history and search exist only on a slashdocs server
- You cannot run and administer your own server: the full mode needs Postgres and S3-compatible storage
- You need to create service accounts or keys through MCP: that is not possible, keys are minted in the web app or CLI
Example request
Write up the incident as an HTML document, publish it to slashdocs and send me the link, then read the comments and publish a new versionLimitations
The main part of the features needs a slashdocs server: you run Postgres and S3-compatible storage and set up OpenID Connect for people to sign in (Google, Okta, Entra ID, Keycloak, Dex and others). The `docker compose up` quickstart brings up a server with Postgres, an S3 gateway and a Dex test provider, and the Dex sign-in is for development only. The app and document rendering must live on different hostnames, or cookie isolation fails.
Uploaded HTML is treated as untrusted and opens in a sandboxed frame. Default sharing for new documents is set on the server through an environment variable, a document is private by default, but an administrator can grant view, comment or edit rights. The `--local` mode has no sharing, versions, full-text search, labels or notifications. The service collects no telemetry. The project was formerly called aidocs, and old links keep working.
How to disable. Delete the skill folder from the agent's skills directory and the `slashdocs` binary from `PATH`. Revoke service account keys in the web app on the server. Documents on the server stay; an administrator removes them.
FAQ
How do I install slashdocs for Claude Code?
The skill and CLI install through the `curl ... | sh` installer from your slashdocs server at `/install.sh`. You can build from source: `git clone https://github.com/razorpay/slashdocs.git`, then `make build`, and the binary appears at `bin/slashdocs`. Standalone CLI archives are on GitHub releases.
Do I need a server to try it?
No. `slashdocs --local` runs on your computer: documents live in `~/slashdocs` and comments are supported. It has no sharing, versions, search, labels or notifications.
Why a single HTML file?
Such a file is easy for an agent to produce, version, archive and move through a CLI. Images are embedded as base64 data URLs, and diagrams are Mermaid, inline SVG or HTML and CSS.
How does the agent get comments?
Through `slashdocs docs comments list <id>` or MCP tools. A comment is anchored to a quote from the text and can be resolved and reopened. The agent reads the edits and publishes a new version with a short change summary.
How do I connect MCP?
Add `https://your-host/mcp` to an MCP client. The client opens the sign-in and consent page and then receives a short-lived token through authorization code with PKCE. Headless agents can use a service account key as a Bearer token.
Is it safe to open HTML written by an agent?
Uploaded HTML is treated as untrusted. The app and document rendering must run on different hostnames, and a document opens in a sandboxed frame, so the browser keeps it away from your session.
Discussion
No questions or reviews yet.
Related
CLI toolsArtifact Share CLI: an agent publishes HTML, Markdown and static sites at a stable link and updates it with new versions
A plugin with a CLI and skills for Claude Code and Codex: agent-made HTML goes to a private AWS archive readable from a phone.
The agent turns rich answers, files, folders and service exports into an HTML page in a ready style, checked in a browser
A skill, an MCP server and a CLI turn Markdown from an agent into a short link to a web page that opens without signing in.
The fl CLI puts a folder of Markdown and HTML online as a site in one command and returns the URL and a claim link to an agent.
A skill plus the docx program: the agent fills forms, edits tables and reviews Word files with tracked changes
README in short
The description from Razorpay calls the project a workspace for agentic organizations and lists the documents you can keep in it. Docker Compose quickstart, build from source, the Docker image and MCP client setup follow. A CLI quickstart, an architecture diagram, development targets and a no-telemetry statement close it.