Skip to content

files.md

Latest commit

 

History

History
210 lines (152 loc) · 6.81 KB

files.md

File metadata and controls

210 lines (152 loc) · 6.81 KB
Raw

Files Connector (CSV, JSON, TXT)

Point MCP-Maker at a directory and it auto-generates tools for every file inside.

No extra installation needed — file support is built into the core mcp-maker package.

Quick Start

mcp-maker init ./my-data/
mcp-maker serve

URI Format

mcp-maker init ./data/              # Relative path to directory
mcp-maker init /absolute/path/data/ # Absolute path

How Files Are Handled

File Type Treated As What You Get
.csv Table list, search, count, schema tools
.json (array of objects) Table list, search, count, schema tools
.txt, .md, .log Resource read tool (returns content)
.yaml, .toml, .xml Resource read tool (returns content)

Example: Directory with Mixed Files

Your directory structure

my-data/
├── customers.csv
├── products.json
├── README.md
└── config.txt

customers.csv

id,name,email,city
1,Alice,alice@acme.com,NYC
2,Bob,bob@globex.com,London
3,Carol,carol@initech.com,Tokyo

products.json

[
  {"sku": "PRD-001", "name": "Widget", "price": 9.99, "category": "Tools"},
  {"sku": "PRD-002", "name": "Gadget", "price": 24.99, "category": "Electronics"},
  {"sku": "PRD-003", "name": "Doohickey", "price": 14.99, "category": "Tools"}
]

Generate the server

mcp-maker init ./my-data/
⚒️ MCP-Maker                                         v0.2.3

  ✅ Connected to files source

  ┌───────────────────────────────────────────────────────┐
  │ 📊 Discovered Tables (2)                             │
  ├───────────┬──────────────────────────┬──────┬────────┤
  │ Table     │ Columns                  │ Rows │ PK     │
  ├───────────┼──────────────────────────┼──────┼────────┤
  │ customers │ id, name, email, city    │    3 │ —      │
  │ products  │ sku, name, price, ...    │    3 │ —      │
  └───────────┴──────────────────────────┴──────┴────────┘

  ┌───────────────────────────────────────────────────────┐
  │ 📄 Discovered Resources (2)                          │
  ├──────────┬───────────────────────────────────────────┤
  │ Name     │ Type                                      │
  ├──────────┼───────────────────────────────────────────┤
  │ readme   │ text/markdown                             │
  │ config   │ text/plain                                │
  └──────────┴───────────────────────────────────────────┘

  🎉 Generated: mcp_server.py

Generated tools

New in 0.2.7:

filter_customers(field, value, operator)   → eq/ne/gt/gte/lt/lte/contains filtering
aggregate_customers(group_by, agg_function, agg_column) → count/sum/avg/min/max
distinct_customers(field)                  → Distinct values of a column
  • TSV and JSONL files are now supported alongside CSV and JSON
  • Single-file mode: mcp-maker init ./users.csv works without a directory
  • Column headers are sanitized (First Name → first_name) and de-duplicated

For CSV/JSON tables:

list_customers(limit=50, offset=0)     → {results, total, has_more, next_offset}
search_customers(query="alice")         → Search across all columns
count_customers()                       → Total row count
schema_customers()                      → Column names & types
export_customers_csv()                  → Export as CSV string
export_customers_json()                 → Export as JSON string

list_products(limit=50, offset=0)      → {results, total, has_more, next_offset}
search_products(query="widget")         → Search products
count_products()                        → Count products
schema_products()                       → Product schema
export_products_csv()                   → Export as CSV string
export_products_json()                  → Export as JSON string

Advanced list features:

list_customers(fields="name,email")     → Column selection
list_customers(sort_field="name")       → Sorting

For text files:

read_readme()     → Returns the content of README.md
read_config()     → Returns the content of config.txt

Ask Claude

You: "What products do we have?"

Claude: calls list_products() "You have 3 products: Widget ($9.99), Gadget ($24.99), and Doohickey ($14.99)."

You: "Find all customers in NYC"

Claude: calls search_customers(query="NYC") "I found 1 customer in NYC: Alice (alice@acme.com)."

You: "Show me the README"

Claude: calls read_readme() Shows the full content of README.md

Column Type Inference

For CSV files, MCP-Maker infers types from the data:

Data Pattern Inferred Type Example
Numbers only Integer 1, 42
Decimal numbers Float 9.99, 3.14
Everything else String "hello"

For JSON files, types come from the JSON:

JSON Type Mapped To
string String
number (int) Integer
number (float) Float
boolean Boolean
array JSON
object JSON

Tips

  • File names become table names. customers.csv → customers table, product-list.json → product_list table.
  • First row of CSV = headers. Make sure your CSV has header names in row 1.
  • JSON must be an array of objects. [{"a":1}, {"a":2}] works. {"a":1} alone is treated as a resource.
  • Nested directories are not recursively scanned (only top-level files).

Troubleshooting

"No tables or resources found"

The directory exists but has no recognizable files.

Fix: Make sure your directory contains .csv, .json, .txt, or .md files.

CSV parsing errors

If your CSV has unusual delimiters or encoding:

Fix: MCP-Maker uses Python's built-in CSV parser. Make sure your file uses commas as delimiters and UTF-8 encoding.