Rulemorph MCP Server

Rulemorph transforms data from external APIs, CSV, JSON, YAML, TOML, XML, HTML, Markdown, and Excel into predictable JSON using declarative YAML/JSON rules.

Instead of adding another custom script for every input source, you can keep transformation behavior in rule files. The same rules can be reused from the CLI, embedded in Rust, served through a local UI/API server, or exposed to AI assistants through MCP.

Try it in your browser: playground.rulemorph.com

What It Solves

Rulemorph moves growing transformation code into reviewable, versioned rules.

• Normalize vendor API responses into your internal schema
• Bring CSV / Excel imports into a JSON pipeline
• Extract values from HTML or XML and process them with the same rule model
• Review, replace, and version transformation behavior as YAML/JSON
• Reuse the same transformation from the CLI, a local API, or an AI assistant

It is not meant to replace application code for arbitrary execution, complex domain logic, or long-running workflow orchestration. For those cases, normal application code or a workflow engine is usually a better fit.

Quick Start

Transform a users array from an external API response into the JSON shape your application expects.

rules.yaml

version: 2
input:
  format: json
  json:
    records_path: "users"
mappings:
  - target: "id"
    source: "user_id"
  - target: "name"
    expr: ["@input.full_name", trim]
  - target: "email"
    expr: ["@input.username", concat: ["lit:@example.com"]]

input.json

{ "users": [{ "user_id": 1, "full_name": " Alice ", "username": "alice" }] }

Run

rulemorph transform -r rules.yaml -i input.json

You can also pipe input to transform:

cat input.json | rulemorph transform -r rules.yaml
cat input.json | rulemorph transform -r rules.yaml -i -

Output

Show output

[{ "id": 1, "name": "Alice", "email": "alice@example.com" }]

For quick one-off transformations without a rule file, use direct mode.

Evaluate a single expression against JSON or ad-hoc CSV:

echo '{ "test": 1 }' | rulemorph -rule '@input.test'
echo '{ "a": 1, "b": 2 }' | rulemorph --rule '["@input.a", {"+": ["@input.b"]}]'
echo 'a,test,1' | rulemorph -rule '@input.0'
echo 'a,test,1' | rulemorph -H 'id,name,age' -rule '@input.id'

Show output

1
3
"a"
"a"

Use -F/--field when you want a small output object and field order matters:

echo 'u1,Alice,42' | rulemorph -H 'id,name,age' \
  -F id='@input.id' \
  -F name='["@input.name","trim","uppercase"]' \
  -F age='["@input.age","int"]'

Show output

{ "id": "u1", "name": "ALICE", "age": 42 }

Use --output-map when a compact nested target map is easier to read:

echo 'u1,Alice,42' | rulemorph -H 'id,name,age' \
  --output-map '{"user.id":"@input.id","user.name":["@input.name","trim"],"age":["@input.age","int"]}'

Show output

{ "user": { "id": "u1", "name": "Alice" }, "age": 42 }

For multi-record direct input, add --ndjson to emit one JSON value per line:

printf 'u1,Alice,42\nu2,Bob,7\n' | rulemorph --ndjson -H 'id,name,age' \
  --output-map '{"id":"@input.id","age":["@input.age","int"]}'

Show output

{"age":42,"id":"u1"}
{"age":7,"id":"u2"}

Shortened. The full README is on GitHub.