DOAJ Discovery MCP

Search the Directory of Open Access Journals from any AI client — no account, no API key, no payment

Problem · Tools · Quick Start · Local · Docker · Privacy · Development · Deployment

🎯 The Problem

The DOAJ indexes over 23,000 peer-reviewed open access journals, but finding the right one is manual work. Its API speaks Elasticsearch query syntax — bare multi-word searches are AND-ed together, so a natural-language question or a pasted abstract silently returns nothing. Filtering by article processing charge, licence, publisher country, or language means knowing field paths like bibjson.apc.has_apc and that DOAJ stores "NL" rather than "Netherlands".

Meanwhile, an AI assistant asked "which no-fee journals publish Syriac manuscript studies?" will answer from training data — plausible journal names, stale APC policies, no verifiable links.

This server closes that gap. It translates plain-language questions and manuscript abstracts into valid DOAJ queries, applies real metadata filters server-side, ranks candidates locally, and returns source links for verification.

Related MCP server: Earthdata MCP Server

🧰 Tools

Eight read-only tools. None performs editorial review, acceptance prediction, compliance checking, or publishing decisions.

Every search response includes the effective DOAJ query it ran, the upstream total, and how many results were returned, so you can see and audit what was actually asked.

Why connect this instead of asking directly?

An AI client can query current DOAJ metadata rather than relying on model memory. It can find journal or article candidates, identify no-APC records, rank results for a topic, and preserve links for verification. The connection exposes no private DOAJ data.

🚀 Quick Start

Remote (Streamable HTTP)

No account, API key, or payment required. Add this URL:

https://doaj-discovery-mcp-hbyczavkfq-ew.a.run.app/mcp

Client availability and menu names change. See the current Claude connector guide and ChatGPT developer mode guide.

Local (stdio)

For clients that launch MCP servers as a subprocess:

npm ci && npm run build

{
  "mcpServers": {
    "doaj-discovery": {
      "command": "node",
      "args": ["/absolute/path/to/doaj-discovery-mcp/dist/src/stdio.js"]
    }
  }
}

💻 Run Locally

Requires Node.js 22 or newer.

npm ci
cp .env.example .env
npm run dev

Open http://localhost:3000/. The MCP endpoint is http://localhost:3000/mcp; health is http://localhost:3000/health. For the stdio transport instead, run npm run dev:stdio.

Configuration is documented in .env.example. The public DOAJ API is used without an API key.

🐳 Docker

docker build -t doaj-discovery-mcp .
docker run --rm -p 3000:3000 --env-file .env doaj-discovery-mcp

🔒 Privacy

Do not send confidential, unpublished, personal, or sensitive manuscript text to the public service. Requests pass through Google Cloud and the public DOAJ API. Query text and abstracts are not intentionally persisted by the application, and production caching is disabled. Read PRIVACY.md for details.

🛠 Development

npm run check                 # tests, build, lint, format check
DOAJ_LIVE_TEST=1 npm test     # additionally hit the real DOAJ API
docker build -t doaj-discovery-mcp:local .

The live tests are skipped by default so CI stays hermetic. They exist as a regression guard: a DOAJ query can be syntactically valid and still match zero records, which unit tests with a stubbed fetch cannot detect.

See CONTRIBUTING.md and the MIT License.

☁️ Deployment

Pushes to main are verified by CI and then deployed to Google Cloud Run through keyless Workload Identity Federation. See docs/CLOUD_RUN.md.

• Landing page: https://doaj-discovery-mcp-hbyczavkfq-ew.a.run.app/

• Health: https://doaj-discovery-mcp-hbyczavkfq-ew.a.run.app/health

• Issues: https://github.com/ramiramirez-nl/doaj-discovery-mcp/issues