Sync Licensing MCP Server
A local MCP server for music sync-licensing that enables searching tracks, checking rights clearance, calculating license costs, generating contracts, and registering usage via JSON-RPC 2.0.
README
Sync Licensing MCP Server
A local Model Context Protocol server that exposes the catalog and the business logic of a music sync-licensing platform: searching tracks by creative brief, checking their rights clearance, quoting a licence under conditional pricing rules, issuing the contract and registering the usage.
Built for CC3067 Redes (Universidad del Valle de Guatemala), Project 1. The MCP message flow is implemented directly on top of JSON-RPC 2.0 — no MCP SDK, no FastMCP, no framework. The server package depends on the Python standard library only.
Table of contents
- The business case
- Architecture
- Requirements
- Installation
- Building the catalog
- Usage
- Tool reference
- Pricing rules
- Protocol details
- Where the data comes from
- Testing
- Project layout
- Project status
1. The business case
Sync licensing is the business model of platforms such as Epidemic Sound, Artlist and Musicbed: a creator or an ad agency must buy a licence before using a track in audiovisual content. The process has three frictions:
- Finding a track that fits the creative brief and the budget is slow.
- The legal status of a track is not obvious — it may contain samples that were never cleared, or be frozen by an authorship dispute.
- The price is not fixed. The same track costs one thing for an Instagram post and something else entirely for a national TV campaign.
This server turns that workflow into five tools an assistant can chain. It is not a search engine with a price list attached: the fee is computed from conditional rules, and the tools refuse operations that would put the client at legal risk.
2. Architecture
┌────────────────────────┐
│ Host (chatbot / CLI) │
└───────────┬────────────┘
│ spawns as a subprocess
┌───────────▼────────────┐
│ MCP client │ client/mcp_cli.py
└───────────┬────────────┘
│ JSON-RPC 2.0 over stdio
│ (one JSON object per line)
┌───────────▼────────────┐
│ MCP server │ synclicense_mcp/
│ │
│ jsonrpc.py framing │
│ server.py dispatch │
│ tools.py 5 tools │
│ pricing.py rate card│
│ contracts.py contracts
│ catalog.py catalog │
└───────────┬────────────┘
│
┌───────────▼────────────┐
│ data/catalog.json │ built by scripts/seed_catalog.py
│ data/usage_log.jsonl │ append-only audit log
└────────────────────────┘
stdout carries protocol traffic only; every diagnostic the server prints goes
to stderr, so piping the server's output never corrupts the stream.
3. Requirements
- Python 3.10 or newer (developed on 3.11).
- No other dependency to run the server.
requestsis only needed to pull real metadata from Jamendo, andpytestonly to run the test suite. Both are inrequirements.txt.
4. Installation
git clone https://github.com/ecarcamo/MCP-Local-Redes.git
cd MCP-Local-Redes
python3 -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r requirements.txt
The package is not installed: it is imported from the repository root, so every command below is run from the project directory.
5. Building the catalog
The repository already ships a catalog at data/catalog.json with 800 real
tracks pulled from the Jamendo API, so you can skip this section and go
straight to Usage. Rebuild it only if you want a different size, a
different seed, or a catalog that needs no credentials.
Offline mode (default, no credentials, no network)
python scripts/seed_catalog.py --offline --count 800
Deterministic: the same --seed always produces the same catalog. It also pins
three known tracks at the top (TRK-00001 cleared, TRK-00002 with pending
samples, TRK-00003 blocked), which makes the failure scenarios easy to
demonstrate.
Jamendo mode (real Creative Commons metadata)
Register at https://devportal.jamendo.com to get a client_id, then:
cp .env.example .env
# edit .env and set JAMENDO_CLIENT_ID=your_client_id
python scripts/seed_catalog.py --jamendo --count 800
Track metadata comes from the API; the base fee and the rights status are still
generated locally (see section 10). Popularity
is taken from the API's own popularity_total ordering. The free Jamendo plan
throttles bursts of requests and answers a throttled page with an empty result
list rather than an error, so the script pauses between pages and retries an
empty page before concluding the catalog is exhausted.
| Option | Default | Description |
|---|---|---|
--offline / --jamendo |
--offline |
Source of the track metadata |
--count N |
800 |
How many tracks to write |
--seed N |
23016 |
Seed for the simulated business layer |
--output PATH |
data/catalog.json |
Where to write the catalog |
6. Usage
6.1 Run the guided demo
A scripted end-to-end run, useful as a smoke test. It spawns the server, plays
the complete licensing conversation, and prints every JSON-RPC message that
crosses the wire (--> sent, <-- received):
python client/mcp_cli.py --demo
The demo walks through: handshake → tools/list → search a track → check its
clearance → quote it → issue the contract → register the usage → and three
failure cases (a blocked track, a quote that belongs to another track, and an
invalid argument).
Add --quiet to hide the raw protocol trace and see only the answers:
python client/mcp_cli.py --demo --quiet
6.2 Interactive session (the main way to use it)
A REPL to drive the server by hand, one tool at a time:
python client/mcp_cli.py --interactive
| Command | Description |
|---|---|
list |
Tools published by the server |
schema <tool> |
JSON Schema of one tool |
call <tool> <json> |
Call a tool with JSON arguments |
vars |
Ids remembered from previous answers |
ping |
Send a JSON-RPC ping |
raw <method> [json] |
Send any JSON-RPC method by hand |
quit |
Close the session |
Ids are remembered. Every *_id a tool returns is stored and can be reused
as $name in the next call, so a whole licensing negotiation can be typed
without copying a single id by hand:
mcp> call buscar_pista {"mood": "epico", "instrumental": true, "presupuesto_max": 100, "limite": 3}
...
remembered: $pista_id=TRK-00312
mcp> call verificar_clearance {"pista_id": "$pista_id"}
mcp> call calcular_costo_licencia {"pista_id": "$pista_id", "tipo_uso": "publicidad_online", "territorio": "latam", "exclusividad": "sectorial", "duracion_meses": 12}
...
remembered: $cotizacion_id=COT-719E615733
mcp> call generar_contrato {"pista_id": "$pista_id", "cliente": "Agencia Lumen S.A.", "cotizacion_id": "$cotizacion_id"}
...
remembered: $contrato_id=CTR-F9D1D72B0D
mcp> call registrar_uso {"contrato_id": "$contrato_id", "plataforma": "YouTube", "url_proyecto": "https://youtube.com/watch?v=demo"}
mcp> vars
mcp> quit
$pista_id defaults to the top candidate of the last search. Use vars at any
point to see what is currently remembered.
6.3 Run the server on its own
python -m synclicense_mcp
It then waits for JSON-RPC messages on stdin. Use --catalog PATH to point it
at a different catalog file.
6.4 Talk to it with no client at all
Because the transport is just newline-delimited JSON, you can drive the server straight from the shell:
printf '%s\n' \
'{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-11-25","capabilities":{},"clientInfo":{"name":"shell","version":"0"}}}' \
'{"jsonrpc":"2.0","method":"notifications/initialized"}' \
'{"jsonrpc":"2.0","id":2,"method":"tools/list"}' \
'{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"verificar_clearance","arguments":{"pista_id":"TRK-00001"}}}' \
| python -m synclicense_mcp
7. Tool reference
| Tool | Required arguments | Returns |
|---|---|---|
buscar_pista |
(none — every filter is optional) | Candidate tracks with id, title, artist, duration and base fee |
verificar_clearance |
pista_id |
Legal status: cleared, samples pending, or blocked |
calcular_costo_licencia |
pista_id, tipo_uso, territorio, exclusividad, duracion_meses |
Full fee breakdown, total in USD, and a cotizacion_id |
generar_contrato |
pista_id, cliente, cotizacion_id |
Contract with scope, term, amount, restrictions, and a contrato_id |
registrar_uso |
contrato_id, plataforma, url_proyecto |
Usage record filed for royalties and audit |
7.1 buscar_pista
Optional filters: mood, genero, instrumental, duracion_seg_min,
duracion_seg_max, presupuesto_max, limite (1–20, default 5).
mood:alegre,epico,melancolico,relajado,tenso,energetico,inspirador,oscurogenero:pop,rock,electronica,hip_hop,jazz,clasica,folk,ambient,cinematica,latina
Tracks blocked by an authorship dispute are excluded: they cannot be licensed, so offering them would be a false positive.
7.2 verificar_clearance
| Status | Licensable | Effect |
|---|---|---|
libre |
yes | No encumbrance |
samples_pendientes |
yes | +15% escrow surcharge and a hold-back clause |
bloqueada |
no | Authorship dispute; quoting and contracting are refused |
7.3 calcular_costo_licencia
| Argument | Allowed values |
|---|---|
tipo_uso |
redes_sociales, evento_interno, podcast, web_corporativo, publicidad_online, videojuego, tv_nacional, cine |
territorio |
local, latam, europa, norteamerica, mundial |
exclusividad |
no, sectorial, total |
duracion_meses |
0 (perpetual) or 1–120 |
Example request and response:
--> {"jsonrpc":"2.0","id":5,"method":"tools/call","params":{
"name":"calcular_costo_licencia",
"arguments":{"pista_id":"TRK-00312","tipo_uso":"redes_sociales",
"territorio":"local","exclusividad":"no","duracion_meses":6}}}
<-- {"jsonrpc":"2.0","id":5,"result":{
"content":[{"type":"text","text":"Quote for TRK-00312 \"Stop!\" ... TOTAL USD 94.50"}],
"structuredContent":{
"ok":true,
"cotizacion_id":"COT-3D18B1547D",
"pista_id":"TRK-00312",
"alcance":{"tipo_uso":"redes_sociales","territorio":"local",
"exclusividad":"no","duracion_meses":6},
"desglose":{"tarifa_base_usd":94.5,
"multiplicadores":{"tipo_uso":1.0,"territorio":1.0,
"exclusividad":1.0,"vigencia":1.0},
"subtotal_usd":94.5,"recargo_escrow_usd":0.0,
"total_usd":94.5,"moneda":"USD"},
"valida_hasta":"2026-09-19T18:15:54+00:00"},
"isError":false}}
7.4 Tool chaining
The tools are stateful within a session, which is the point of the use case:
buscar_pista ──► pista_id
├──► verificar_clearance (can stop the whole flow)
└──► calcular_costo_licencia ──► cotizacion_id
└──► generar_contrato ──► contrato_id
└──► registrar_uso
generar_contrato rejects a quote that does not exist, has expired (30 days),
or was issued for a different track. registrar_uso rejects an unknown or
inactive contract. Quotes and contracts belong to one connection and are not
shared between sessions.
8. Pricing rules
subtotal = tarifa_base × mult_use × mult_territory × mult_exclusivity × mult_term
total = subtotal + escrow surcharge (15% when the track has pending samples)
| Type of use | × | Territory | × | Exclusivity | × | Term | × |
|---|---|---|---|---|---|---|---|
redes_sociales |
1.0 | local |
1.0 | no |
1.0 | ≤ 3 months | 0.8 |
evento_interno |
1.1 | latam |
1.8 | sectorial |
2.0 | ≤ 6 months | 1.0 |
podcast |
1.3 | europa |
2.2 | total |
4.5 | ≤ 12 months | 1.5 |
web_corporativo |
1.6 | norteamerica |
2.4 | ≤ 24 months | 2.2 | ||
publicidad_online |
2.5 | mundial |
3.2 | ≤ 36 months | 2.8 | ||
videojuego |
4.0 | > 36 months | 3.2 | ||||
tv_nacional |
6.0 | perpetual | 3.5 | ||||
cine |
8.0 |
Six months is the reference term, which is why it sits at 1.0. A quote holds its price for 30 days.
9. Protocol details
Transport. stdio, one JSON-RPC 2.0 message per line, UTF-8, no embedded newlines. The server exits cleanly on EOF.
Protocol versions. 2025-11-25 (preferred) and 2025-06-18. If the client
asks for anything else, the server answers with its preferred version instead
of failing the handshake.
Methods.
| Method | Result |
|---|---|
initialize |
Negotiated version, capabilities, server info, instructions |
notifications/initialized |
(notification — no response) |
ping |
{} |
tools/list |
The five tool descriptors with their JSON Schemas |
tools/call |
content, structuredContent, isError |
Error codes.
| Code | Meaning |
|---|---|
-32700 |
Parse error — the line is not valid JSON |
-32600 |
Invalid request — bad envelope |
-32601 |
Method not found |
-32602 |
Invalid params — missing, ill-typed or out-of-enum argument, or unknown tool |
-32603 |
Internal error |
-32002 |
Server not initialized — a request arrived before the handshake |
Protocol errors vs. business errors. A malformed call comes back as a
JSON-RPC error. A well-formed call that the licensing rules refuse — a blocked
track, an expired quote, an unknown contract — comes back as a successful
response carrying isError: true and a readable explanation, so a model can
read the reason and correct course instead of seeing a transport failure.
A full specification is in docs/SERVER_SPEC.md.
10. Where the data comes from
Track metadata (title, artist, duration, genre, mood, licence, popularity ranking) comes from the public Jamendo API, which exposes a Creative Commons catalog. The catalog shipped in this repository was built that way. The offline generator produces the same shape locally, so the project still runs with no credentials and no network access.
The business layer is simulated, on purpose. No platform publishes its rate
card or the internal legal status of each track, so tarifa_base_usd and
estado_derechos are generated from a fixed seed with a realistic distribution
(82% cleared, 13% samples pending, 5% blocked). The rate-card multipliers were
designed from the public royalty-free rate cards of platforms such as
Jamendo Licensing.
This scope was reviewed and approved by the course instructor.
11. Testing
python -m pytest tests/ -v
The suite covers the rate-card rules, the JSON-RPC framing, the handshake, the error codes, the tool chain and its refusals, the seed generator, and one end-to-end test that launches the real server process and speaks the stdio transport to it. The tests look tracks up by rights status rather than by a fixed id, so they pass against any catalog: offline, Jamendo, or regenerated with a different seed.
12. Project layout
MCP-Local-Redes/
├── synclicense_mcp/ MCP server package (standard library only)
│ ├── __main__.py entry point: python -m synclicense_mcp
│ ├── jsonrpc.py JSON-RPC 2.0 framing over stdio
│ ├── server.py MCP method dispatch
│ ├── tools.py the five tools: schemas, validation, handlers
│ ├── pricing.py conditional rate card
│ ├── contracts.py contracts and usage registration
│ ├── catalog.py catalog loading and search
│ └── errors.py business-rule failures
├── client/mcp_cli.py manual JSON-RPC client (demo + REPL)
├── scripts/seed_catalog.py catalog builder (offline / Jamendo)
├── data/catalog.json generated catalog
├── tests/ pytest suite
└── docs/ proposal, assignment brief, server specification
13. Project status
Delivered in this stage:
- Local MCP server over stdio with the five tools of the approved use case.
- JSON-RPC 2.0 and the MCP handshake implemented by hand.
- Command-line client with a scripted demo and an interactive REPL.
- Catalog seeding, in both offline and Jamendo modes.
- Test suite.
Planned for the rest of the project:
- Chatbot host on the Anthropic API, with session context and a visible log of every MCP interaction.
- Integration with the official Filesystem and Git MCP servers.
- The same server deployed remotely over HTTP.
- Wireshark capture and layer-by-layer analysis of the remote traffic.
Author: Esteban Cárcamo (23016) — CC3067 Redes, Section 20
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。