Quiver

Finally, marketing that makes sense to engineers.

Quiver is an open-source, self-hosted agentic developer marketing system for technical founders and teams building developer tools. It connects product context, customer research, campaigns, content, and results, so the next piece of work starts with what your team knows—not another empty chat. Work in the web app or through MCP-compatible clients; keep the history and decisions in Quiver.

Choose your path: Self-host this repository · Use hosted Quiver · Contribute

The problem isn't another draft

Your positioning lives in one document. Customer objections are buried in call notes. Campaign work happens in a chat; published content lives somewhere else; results rarely make it back into the next brief. A capable agent can still produce the wrong work when it cannot see the decisions behind it.

Quiver gives that work a system of record. Product context has versions. Research can inform campaigns. Sessions produce artifacts you can review and revise. Published content and performance stay linked to the work that produced them. When results suggest a change in direction, the team can review a proposed context update rather than silently rewriting its source of truth.

A concrete loop

Say you're marketing a new developer API:

1. Set the context. Record the product, audience, positioning, messaging, and hypotheses in onboarding. Quiver keeps versions as that context changes.
2. Capture evidence. Save customer research and feature the quotes worth carrying into future strategy and creation sessions.
3. Run the campaign. Use Strategy to plan and Create to draft content. Save the outputs as versioned artifacts linked to a campaign; move them through review before calling them live.
4. Ship and learn. Track content and where it was distributed, log results against the campaign or artifact, and review proposed context changes before the next round.

The web UI runs Quiver's five AI session modes. MCP clients can work with the same context, campaigns, artifacts, research, content, and performance records after initial setup in the web UI. Quiver organizes the work; it does not replace your judgment, your publishing destination, your CRM, or your analytics tools.

What's in the self-hosted edition

Part	What it lets you do
Product context	Keep approved positioning, ICP, messaging, evidence, and hypotheses in versioned context; review proposals and restore earlier versions.
Sessions & skills	Use Strategy, Create, Feedback, Analyze, or Optimize modes with skills loaded for the task and product context available to each session.
Campaigns & artifacts	Group sessions, research, content, artifacts, and results around an initiative. Save revisions and move artifacts through explicit statuses.
Customer research	Save research entries, extract and feature customer quotes, and capture signals against active hypotheses.
Content	Store markdown with SEO/OG metadata; track distributions and dated metric snapshots. Make published pieces available to your site through the public Content API.
Performance	Log metrics and qualitative results; review the close-the-loop queue and proposed context updates.
MCP	Use Quiver from an external agent with the local stdio server or remote HTTP endpoint.

Quiver is not an AI copy generator with a folder bolted on. Its value is keeping the source material, the work, its state, and what happened afterward connected. It also isn't a turnkey autopublisher: the Content API makes published records available for your site to fetch and render.

Self-hosting

This repository is the MIT-licensed, single-team deployment. You operate the app, database, and Anthropic account. If you'd rather have a managed, shared team workspace without maintaining the infrastructure, use hosted Quiver instead; its packaging and features are separate from this repository.

You'll need a Supabase project (Postgres and Auth), an Anthropic API key, and a Vercel project. The deploy button creates a deployment, but you still need to configure the services and run the database migrations.

1. Fork this repository. Create a Supabase project and collect its connection strings, project URL, and API keys.
2. Generate a share-link secret with openssl rand -base64 32. Set the environment variables on your Vercel project; configure Supabase Auth for your deployment URL.
3. Use the Deploy with Vercel button above, or import your fork into Vercel. Set NEXT_PUBLIC_APP_URL to your actual deployment URL.
4. From your checked-out fork with DATABASE_URL and DIRECT_URL configured, apply the schema:

   npx prisma migrate deploy

5. Open the deployment. A fresh install offers Create the first account on the login screen — that account becomes the admin. Complete onboarding to create the first product-context version.

For local development, see CONTRIBUTING.md. Don't commit .env.local or expose the Supabase service-role key in client-side code.

Environment variables

Copy .env.example to .env.local for local work, or set the same keys in Vercel. The example file explains where to obtain each value.

Variable	Required	Purpose
DATABASE_URL	Yes	Pooled Postgres connection string for the app.
DIRECT_URL	Yes	Direct Postgres connection for Prisma migrations.
NEXT_PUBLIC_SUPABASE_URL	Yes	Supabase project URL.
NEXT_PUBLIC_SUPABASE_ANON_KEY	Yes	Supabase public/anon key.
SUPABASE_SERVICE_ROLE_KEY	Yes	Server-only service-role key. Never expose it to clients.
ANTHROPIC_API_KEY	Yes	Anthropic API key for the app's AI features.
NEXT_PUBLIC_APP_URL	Yes	Your app URL, e.g. https://your-domain.example.
QUIVER_SHARE_SECRET	Yes	Secret for session share links; generate it yourself.
MCP_AUTH_SECRET	Yes in production	Bearer token for /api/mcp. Without it, that endpoint refuses every request outside local development — it exposes writes and deletes.
CRON_SECRET	Yes in production	Bearer token Vercel sends to /api/cron/pattern-report. Without it the job returns 200 and does nothing, so the monthly pattern report never runs.

Public Content API

Your site can fetch published content from Quiver as JSON, then render it however you want:

GET /api/public/content/[slug]   # one published piece, including markdown and SEO/OG fields
GET /api/public/content          # paginated list of published pieces

The list accepts contentType, limit (default 20, max 50), and offset (default 0). Both endpoints are public and rate-limited to 60 requests/minute per IP, using an in-memory limiter per app instance. Drafts are not returned. See app/api/public/content for the implementation.

MCP server

Quiver exposes its context, campaigns, artifacts, content, research, and performance as tools. Connect an MCP-compatible client to work with the same records without opening the web UI—for example, save research after a call or log campaign results from an agent that also has access to your analytics tools. Those external data pulls depend on the integrations your client has; Quiver does not automatically connect to them.

Local stdio server

Build the server from the repository root:

npm install          # repository root first
cd mcp
npm install
npm run build        # runs prisma generate itself

Point Claude Desktop, Cursor, or another stdio MCP client at the absolute path to mcp/dist/index.js and supply the database connection string. Example configuration:

{
  "mcpServers": {
    "quiver": {
      "command": "node",
      "args": ["/absolute/path/to/quiver/mcp/dist/index.js"],
      "env": {
        "DATABASE_URL": "your-postgres-connection-string"
      }
    }
  }
}

ANTHROPIC_API_KEY is optional for this stdio server. Without it, log_performance still stores results but skips AI synthesis.

Remote HTTP endpoint

The Next.js app exposes Streamable HTTP at https://<your-domain>/api/mcp. Set MCP_AUTH_SECRET before deploying. Without it the endpoint refuses every request outside local development, because it exposes tools that write and delete data. Clients then send Authorization: Bearer <your-secret>. Treat the secret and database connection string as credentials.

The tools are grouped by context, campaigns, artifacts, performance, content, research, sessions, and workspace; see mcp/tools for the exact list. propose_context_update queues a change for review. apply_context_update changes the active context immediately and creates a version: use it only when a human explicitly directs the change. MCP access is powerful; connect only clients you trust.

How sessions use context

Each session assembles its prompt from the active product context, skills for the selected mode, and task-specific evidence. Create can include recent performance for that artifact type; Create and Strategy can include featured research quotes and recent published content. See lib/ai/session.ts and lib/ai/skills.ts. The pinned marketing skills live in /skills with upstream attribution; they are not fetched on every request.

Built with

Next.js 14 · TypeScript · Tailwind CSS and shadcn/ui · Supabase/Postgres · Prisma · Anthropic SDK · Vitest · Vercel

Contributing & license

Issues and pull requests are welcome. See CONTRIBUTING.md for local setup, tests, and adding skills; SPEC.md for the product model. Quiver's source is MIT licensed. The vendored skills/ directory is pinned from marketingskills under its own MIT license.