Agent-native structured data format
TMD is a markdown-based format for storing structured, tabular data that is:
- Human-readable — Edit in any text editor
- Git-friendly — Clean diffs, easy merges
- Token-efficient — Optimized for LLM context
- Agent-native — Live queries, cross-file references
---schema:
id: int (pk)name: textamount: decimalpaid: boolcomputed:
total: SUM(amount)outstanding: SUM(amount WHERE paid=false)---# Invoices
invoices[1]: name=Acme Corp, amount=1500.00, paid=false
invoices[2]: name=Widget Inc, amount=750.50, paid=true
## Notes
Follow up with [[contacts/john-smith]] about payment.pip install -e .# Parse and inspect
tmd parse invoices.tmd
# Query data
tmd query invoices.tmd "SUM(amount WHERE paid=false)"
tmd query invoices.tmd "WHERE paid=false"# Compute all formulas
tmd compute invoices.tmd
# Validate
tmd validate invoices.tmd
# Row-safe agent operations
tmd rows invoices.tmd --json
tmd row invoices.tmd 1 --json
tmd append-row invoices.tmd --set name=Northwind --set amount=500 --set paid=false --json
tmd update-row invoices.tmd 1 --set paid=true --json
# Export
tmd export invoices.tmd --format json
tmd export invoices.tmd --format csvTMD includes a dependency-free MCP stdio server for row-aware structured data tools. Point it at a folder that contains .tmd files:
tmd mcp --root ~/Documents/project-orionClaude Desktop config example:
{
"mcpServers": {
"tmd": {
"command": "tmd",
"args": ["mcp", "--root", "/Users/YOU/Documents/project-orion"]
}
}
}Available MCP tools:
tmd_list_files— list.tmdfiles under the configured roottmd_parse_file— parse schema and rowstmd_query_file— run TMD queries likeSUM(amount WHERE paid=false)tmd_get_row— fetch one row by idtmd_append_row— append a row safelytmd_update_row— update fields on an existing rowtmd_validate_file— return validation errors/warnings
TMD files do not need separate vaults. They can live inside existing ContextFit vaults for search/retrieval, while TMD CLI/MCP handles row-aware reads and writes.
fromtmdimportparse_file, query, compute, append_row, update_row# Load documentdoc=parse_file("invoices.tmd")
# Access rowsforrowindoc:
print(row.id, row["name"], row["amount"])
# Filterunpaid=doc.filter(paid=False)
# Querytotal=query(doc, "SUM(amount)")
unpaid_rows=query(doc, "WHERE paid=false")
# Compute all formulasresults=compute(doc)
print(results["outstanding"])
# Row-safe writesappend_row("invoices.tmd", {"name": "Northwind", "amount": 500, "paid": False})
update_row("invoices.tmd", 1, {"paid": True})- Schema with types:
int,decimal,text,bool,date,datetime - References: Wiki-style links
[[path/to/file]] - Computed fields:
SUM(),AVG(),COUNT(),MIN(),MAX() - Filters:
WHERE field=value AND field2>10 - Multiline text: YAML-style blocks for long content
- Prose sections: Mix data with documentation
See SPEC.md for full specification.
MIT