franchise-mcp-server
Enables LLMs to read and edit Madden NFL franchise saves and EA SPORTS College Football dynasty saves (CFB 27+) through 14 well-guarded tools.
README
franchise-mcp-server
An MCP (Model Context Protocol) server that lets LLMs read and edit Madden NFL franchise saves and EA SPORTS College Football dynasty saves (CFB 27+). It wraps bep713/madden-franchise — the community-standard parser behind tools like Franchise Editor — and exposes its full capability as 14 well-guarded tools.
Supported games (inherited from the library): Madden 19–27, College Football 27+. Both franchise saves and FTC (franchise-common) game-data files open.
┌────────────┐ MCP (stdio) ┌─────────────────────┐ ┌──────────────────┐
│ LLM client │ ◄─────────────────► │ franchise-mcp-server │ ◄────► │ madden-franchise │ ◄── CAREER-* / DYNASTY-* files
└────────────┘ └─────────────────────┘ └──────────────────┘
Layout
franchise-mcp-server/
├── src/ TypeScript source
│ ├── index.ts server + tool registrations
│ ├── session.ts open-file sessions, table resolution, field loading
│ ├── serialize.ts field value ⇄ JSON conversion, validation, pagination
│ ├── types.ts structural types for the library
│ └── constants.ts
├── dist/ built JS (entry: dist/index.js)
├── tests/smoke-test.mjs end-to-end test (drives the server over stdio)
└── skills/franchise-editor/ ← docs for LLM agents
├── SKILL.md main skill (workflow + safety rules)
└── references/
├── domain-model.md file anatomy, references, free-lists
├── workflows.md step-by-step recipes
├── cfb-notes.md CFB 27 dynasty specifics
└── library-api.md building apps on the library directly
Setup
Requires Node.js >= 22.19.0.
This server depends on madden-franchise via file:../madden-franchise — clone it
as a sibling directory (the repo version includes CFB 27 support that is
ahead of the npm release):
# 1. Clone both repos side by side
git clone https://github.com/bep713/madden-franchise.git
git clone https://github.com/<you>/franchise-mcp-server.git
# 2. Build the library
cd madden-franchise && npm install && npm run build
# 3. Build the server
cd ../franchise-mcp-server && npm install && npm run build
# 4. Smoke test against bundled CFB 27 test data (uses a copy!)
cp ../madden-franchise/tests/data/DYNASTY-27COMPRESS /tmp/DYNASTY-TEST
node tests/smoke-test.mjs /tmp/DYNASTY-TEST
Register with Claude Code (project .mcp.json)
{
"mcpServers": {
"franchise": {
"command": "node",
"args": ["<absolute-path-to>/franchise-mcp-server/dist/index.js"]
}
}
}
Register with Claude Desktop (claude_desktop_config.json)
{
"mcpServers": {
"franchise": {
"command": "node",
"args": ["<absolute-path-to>/franchise-mcp-server/dist/index.js"]
}
}
}
Give the agent the skills/franchise-editor/ folder (Claude Code discovers it
from .claude/skills/; other agents can be pointed at SKILL.md directly).
Tool catalog
File lifecycle
| Tool | Purpose |
|---|---|
franchise_open |
Open a CAREER-/DYNASTY- file into a session. Options: auto_unempty (needed to add rows), game_year_override, game_type_override. Returns game/schema info. |
franchise_save |
Persist changes — overwrite, or output_path to save-as. Nothing is written to disk without this. |
franchise_close |
Close a session, discarding unsaved changes. |
franchise_list_open_files |
Sessions + which tables have unsaved changes. |
Discovery
| Tool | Purpose |
|---|---|
franchise_list_tables |
Browse/filter the ~2,000+ tables (name, uniqueId, record count, capacity, isArray). |
franchise_get_table_schema |
Field definitions: type, min/max, maxLength, enum members, reference-ness, storage section. |
Reading
| Tool | Purpose |
|---|---|
franchise_read_records |
Paginated row reads with field selection; row_indices for specific rows; array tables return ordered reference lists. |
franchise_get_record |
Every field of one row, field-paginated (Player has ~300 fields). |
franchise_search_records |
Find rows by field conditions (eq/neq/contains/gt/gte/lt/lte), e.g. player by name. |
Writing
| Tool | Purpose |
|---|---|
franchise_edit_records |
Batch field edits with validation: range-checked ints, length-checked strings, enum-name checking, references as {tableId,row}/{tableUniqueId,row}/null, JSON blobs (CharacterVisuals). Echoes old→new values. Blocks writes to empty records unless auto_unempty. |
References & row lifecycle
| Tool | Purpose |
|---|---|
franchise_resolve_reference |
Decode 32-bit refs (binary string, asset id, or tableId+row) to table name/uniqueId/row. |
franchise_find_references_to_record |
Which tables point at a record (run before removing/repurposing anything). |
franchise_get_empty_records |
Free-list state: nextRecordToUse, empty-row chain — the safe way to add rows. |
franchise_recalculate_empty_records |
Rebuild the free-list after bulk operations. |
Design notes
- Validation-first writes. The underlying library bit-packs values and will silently wrap out-of-range ints and corrupt free-lists if misused; this server rejects out-of-range/oversized/invalid-enum writes and empty-record writes with errors that state the valid values.
- Stable identity. Tools push agents toward
table_unique_id(stable across files and game years) over file-specifictableIds and ambiguous names. - Context-efficient. Everything is paginated and field-selectable;
responses over ~25k chars truncate with explicit
truncated_by_size+next_offsetmetadata. - Edit-safe partial reads. Field lists load incrementally; pending edits are flushed to the table buffer before any re-read, so partial reads never lose unsaved changes.
- Sessions. Multiple files can be open at once (e.g. copy data between a Madden franchise and a CFB dynasty, or between a save and its backup).
Known table uniqueIds
| Table | uniqueId | Verified |
|---|---|---|
| Player | 1612938518 |
Madden 25/26, CFB 27 |
| CharacterVisuals | 1429178382 |
CFB 27 |
Everything else: discover per-file with franchise_list_tables (name collisions
are common — eight tables named Team exist in CFB 27; pick by record count).
License
MIT (the wrapped madden-franchise library is MIT by matthewpanetta/bep713).
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。