Ganjoor MCP Server

#

)

English | فارسی

Ganjoor MCP Server — the Model Context Protocol server for the Ganjoor Persian poetry API.

It lets an agent browse poets and collections, read poems with their verses, run prosody and rhyme analysis, trace reply-poems across poets, look up recitations and commentary, and explore the related-people and family-tree graphs.

44 tools, read-only access. No API key required.

---

Contents

• Install
• Docker
• Configuration
• Concepts
• Tool reference
• Response formats
• Usage
• Architecture
• Development
• Known limitations
• Evaluation
• Contributing

---

Install

Requires Node.js 18.17 or newer.

npm install ganjoor-mcp-server # for locally installed stdio transport. no need if using VS Code or Claude Desktop / Cursor.

VS Code

No need install npm package — the MCP server runs in-process and is launched by VS Code when needed.

Just click the badge below to install the Ganjoor MCP server in your VS Code workspace. It will automatically add the server to your .vscode/mcp.json config.

Or add to .vscode/mcp.json:

{

"servers": { "io.github.ehsanghaffar/ganjoor-mcp-server": { "command": "npx", "args": [ "-y", "ganjoor-mcp-server" ], "env": {} } } }

VS Code CLI:

code --add-mcp '{\"name\":\"io.github.ehsanghaffar/ganjoor-mcp-server\",\"command\":\"npx\",\"args\":[\"-y\",\"ganjoor-mcp-server\"],\"env\":{}}'

Claude Desktop / Cursor

Just click the badge below to install the Ganjoor MCP server in your Claude Desktop or Cursor app. It will automatically add the server to your MCP config.

Or add to your MCP config file:

{

"mcpServers": { "io.github.ehsanghaffar/ganjoor-mcp-server": { "command": "npx", "args": [ "-y", "ganjoor-mcp-server" ], "env": {}, "type": "local", "tools": [ "*" ] } } }

Other clients

Claude Code

Use the Claude Code CLI to add the io.github.ehsanghaffar/ganjoor-mcp-server MCP server:

claude mcp add io.github.ehsanghaffar/ganjoor-mcp-server npx -y ganjoor-mcp-server

Claude Desktop

Follow the MCP install guide, use the standard config above.

Codex

Create or edit the configuration file ~/.codex/config.toml and add:

[mcp_servers.ganjoor-mcp-server]

command = "npx" args = ["-y", "ganjoor-mcp-server"]

For more information, see the Codex MCP documentation.

Gemini CLI

Follow the MCP install guide, use the standard config above.

OpenCode

Follow the MCP Servers documentation. For example in ~/.config/opencode/opencode.json:

{

"$schema": "https://opencode.ai/config.json", "mcp": { "io.github.ehsanghaffar/ganjoor-mcp-server": { "type": "local", "command": [ "npx", "-y", "ganjoor-mcp-server" ], "enabled": true } } }

</details>

GitHub Copilot CLI

GitHub Copilot CLI supports adding MCP servers interactively and through a config file.

Option 1: Interactive setup with /mcp add

1. Open GitHub Copilot CLI in interactive mode.
2. Run /mcp add.
3. Enter Server Name: io.github.ehsanghaffar/ganjoor-mcp-server.
4. Choose Server Type:

• STDIO for command-based servers
• HTTP for remote servers

1. Fill in command/url settings from the standard config above and set Tools to *.
2. Press Ctrl+S to save.

Option 2: Edit ~/.copilot/mcp-config.json

{

"mcpServers": { "io.github.ehsanghaffar/ganjoor-mcp-server": { "command": "npx", "args": [ "-y", "ganjoor-mcp-server" ], "env": {}, "type": "local", "tools": [ "*" ] } } }

For more information, see the GitHub Copilot CLI MCP documentation.

GitHub Copilot Coding Agent

GitHub Copilot Coding Agent can use MCP servers to extend its capabilities. Use the configuration below specifically formatted for the Coding Agent:

{

"mcpServers": { "ganjoor-mcp-server": { "command": "npx", "args": [ "-y", "ganjoor-mcp-server" ], "env": {}, "type": "local", "tools": [ "*" ] } } }

Add this configuration to your repository settings under Copilot > Coding agent. The "tools": ["*"] setting enables all available tools from the MCP server.

For more information, see the GitHub Copilot Coding Agent MCP documentation.

Install from this source

npm install

npm run build npm start # stdio transport (default)

VS Code

Add to .vscode/mcp.json:

{

"servers": { "ganjoor": { "type": "stdio", "command": "node", "args": ["${workspaceFolder}/dist/index.js"] } } }

Claude Desktop / Cursor

Add to your MCP config file:

{

"mcpServers": { "ganjoor": { "command": "node", "args": ["/absolute/path/to/ganjoor-mcp/dist/index.js"] } } }

Streamable HTTP (remote / shared deployment)

HTTP=true PORT=3000 npm start

The MCP endpoint is POST /mcp, and GET /health reports server and upstream status. The HTTP transport runs statelessly (a fresh server instance per request), so it scales horizontally. It binds 127.0.0.1 by default and validates the Origin header to block DNS-rebinding attacks; set ALLOWED_ORIGINS (comma-separated) if you need non-loopback origins.

MCP Inspector

npm run inspector

---

Docker

A multi-stage Dockerfile and a docker-compose.yml are included. The image is built on node:22-alpine and contains only the compiled output and production dependencies.

docker compose up --build

The compose file starts the server in streamable HTTP mode on http://localhost:3000/mcp, with GET /health reporting server and upstream status. HOST is set to 0.0.0.0 so the published port is reachable from the host; every configuration variable from the table below works as an ordinary environment variable. To run a single container without compose:

docker build -t ganjoor-mcp .

docker run --rm -p 3000:3000 -e HTTP=true -e HOST=0.0.0.0 ganjoor-mcp

stdio transport over Docker

Clients that launch the server themselves (rather than connecting over HTTP) can use the stdio profile, which opens stdin for the protocol stream and starts no HTTP listener:

docker compose --profile stdio run --rm ganjoor-mcp-stdio

With Claude Desktop

To run the server through Docker instead of a local Node install:

{

"mcpServers": { "ganjoor": { "command": "docker", "args": ["run", "--rm", "-i", "ganjoor-mcp"] } } }

---

Configuration

All settings are environment variables; none are required.

---

Concepts

Ganjoor's data model has three nested levels. Getting these straight is the difference between a productive and a fruitless agent session.

graph TD

P["Poet — e.g. Hafez, id 2, slug /hafez"] --> C1["Category — e.g. Ghazals, id 24"] P --> C2["Category — e.g. Qeta, id 25"] C1 --> M1["Poem — e.g. Ghazal 494, id 2623, /hafez/ghazal/sh494"] C2 --> M2["Poem"] M1 --> V["Verses (hemistichs, paired into couplets)"] M1 --> S["Sections (nested stanzas with metre + rhyme)"] M1 --> R["Recitations (audio)"] M1 --> CM["Comments (reader commentary)"] M1 --> Q["Quoted / related poems"]

• Poet — an author. Hafez is poet id 2, Saadi 7, Rumi 5, Khayyam 3, Ferdowsi 4. ganjoor_list_poets resolves a name to an id.
• Category — a book or section inside a poet's work (CatType 0 general, 1 poems, 2 book). Hafez's root category is 9; his ghazals are 24.
• Poem — one text, identified by a numeric id and a URL slug such as hafez/ghazal/sh494.
• Verse — a hemistich. Two consecutive verses form a couplet (versePosition 0 and 1).

URL slugs need a leading slash. The upstream API returns 404 for ?url=hafez/ghazal/sh494 but

resolves ?url=/hafez/ghazal/sh494. This server normalizes for you, so either form works.

Persian text is returned verbatim. It is right-to-left with Arabic-script diacritics used for

prosody. Do not transliterate or translate it unless the user explicitly asks.

---

Tool reference

Poets and collections

Categories

Poems

| `ganjoor_search_