Structured Data

LocalData MCP handles JSON, YAML, and TOML files as navigable trees, and XML and INI files as flat SQL tables. The connection interface is the same for all five formats; the storage model determines which tools are available after connection.

Tree storage (JSON, YAML, TOML)

When connected, these files are parsed into a tree where each node is individually addressable. Nodes can hold key-value properties and any number of child nodes. An LLM agent can navigate, query, edit, and export the tree without loading the entire file into context.

connect_database("cfg",       "json", "./config.json")
connect_database("settings",  "yaml", "./settings.yaml")
connect_database("pyproject", "toml", "./pyproject.toml")

The connection response includes a tree summary: root nodes, maximum depth, and total property count.

Tree tools

Tool	Description
get_node(name, path)	View a node’s properties and whether it has children
get_children(name, path)	List child nodes with pagination
set_node(name, path)	Create a node; missing ancestors are created automatically
delete_node(name, path)	Remove a node and all its descendants
list_keys(name, path)	List key-value pairs at a node
get_value(name, path, key)	Read a single property value
set_value(name, path, key, value)	Create or update a property; type is inferred automatically
delete_key(name, path, key)	Remove a property from a node
export_structured(name, format)	Export the tree as JSON, YAML, or TOML

The path argument uses dot notation to address nodes: "server" addresses a top-level node, "server.tls" addresses a child node named tls under server.

Navigation

connect_database("cfg", "toml", "./config.toml")

get_node("cfg")               # Root summary: node count, depth, property count
get_node("cfg", "server")     # Properties of the "server" node and child count
get_children("cfg", "server") # Paginated list of child nodes
get_value("cfg", "server", "port")  # → 8080

Editing

set_value("cfg", "server", "port", "9090")

# Dot-notation path; "monitoring" and "alerts" are created if absent
set_value("cfg", "monitoring.alerts", "enabled", "true")

delete_key("cfg", "server", "deprecated_setting")
delete_node("cfg", "monitoring.alerts")  # Removes the node and all children

Format conversion

export_structured converts the in-memory tree to any of the three supported formats, regardless of the source format:

connect_database("cfg", "toml", "./config.toml")
export_structured("cfg", "json")   # Emit as JSON
export_structured("cfg", "yaml")   # Emit as YAML

This lets an agent convert between TOML, JSON, and YAML without writing any conversion logic.

Type inference

set_value infers the type of the supplied string before storing it:

• "true" / "false" → boolean

• Numeric strings → integer or float

• "null" → null

• Anything else → string

• Lists and dicts must be supplied as JSON-serialized strings; they are stored as-is and returned in that form.

XML and INI (flat handling)

XML and INI files connect with the same call but are stored as flat SQL tables rather than trees. After connection, use the standard database tools (execute_query, describe_table, and so on).

connect_database("config_xml", "xml",  "./config.xml")
connect_database("app_ini",    "ini",  "./app.ini")

Both land in a single table named data_table.

XML table layout

The file is flattened with pandas read_xml: each direct child of the root element becomes a row, and the columns come from that child’s attributes and its own sub-elements. Column names have spaces and hyphens replaced by underscores. There is no id, tag, text or parent_id column, and attribute names are not prefixed.

For this file:

<root>
  <item enabled="true">alpha</item>
  <item enabled="false">beta</item>
</root>

the table has columns enabled (from the attribute) and item (from the element text), two rows:

execute_query("config_xml", "SELECT item FROM data_table WHERE enabled = 1")

Deeply nested XML does not flatten cleanly. Run describe_table("config_xml", "data_table") after connecting to see the columns you actually got.

INI table layout

Each key-value pair becomes a row:

Column	Content
section	Section name from the INI file
key	Key within that section
value	Value as a string

execute_query("app_ini", "SELECT value FROM data_table WHERE section = 'database' AND key = 'host'")

Available tools for XML and INI

Because these formats map to SQL tables, all standard database tools apply:

• describe_database(name) — list tables and row counts

• describe_table(name, table) — column names and types

• execute_query(name, sql) — run any read query

• find_table(table_name) — report which connections hold a table of that name; it searches every active connection and takes no connection argument

The tree tools (get_node, get_value, etc.) are not available for XML or INI connections.