@asobacloud/odse-mcp
MCP server for ODS-E energy telemetry conversion. AI coding/data agents call its tools over the Model Context Protocol to convert OEM payloads into ODS-E records and validate them via the published odse Python package.
The installed odse package is the source of truth for OEM support, transform behavior, and validation semantics.
Quick start
Set ODSE_WORKSPACE to an absolute path used when resolving relative payload_file values. If omitted, the server uses its process working directory. The server speaks MCP over stdio (no HTTP port).
Prerequisite: python3 on PATH must be able to import odse (pip install odse).
Cursor
Add to ~/.cursor/mcp.json or the project .cursor/mcp.json:
{
"mcpServers": {
"odse": {
"command": "npx",
"args": ["-y", "@asobacloud/odse-mcp"],
"env": {
"ODSE_WORKSPACE": "/absolute/path/to/your/data"
}
}
}
}
Restart Cursor (or reload MCP servers), then ask the agent to convert or validate ODS-E data.
Claude Code
CLI (user scope):
claude mcp add --transport stdio --scope user \
--env ODSE_WORKSPACE=/absolute/path/to/your/data \
odse -- npx -y @asobacloud/odse-mcp
Or put the same JSON under mcpServers in project .mcp.json (team-shared) or ~/.claude.json (user-wide):
{
"mcpServers": {
"odse": {
"command": "npx",
"args": ["-y", "@asobacloud/odse-mcp"],
"env": {
"ODSE_WORKSPACE": "/absolute/path/to/your/data"
}
}
}
}
Verify with claude mcp list. Project .mcp.json servers need approval the first time you open the repo in Claude Code.
Codex
CLI:
codex mcp add odse --env ODSE_WORKSPACE=/absolute/path/to/your/data -- npx -y @asobacloud/odse-mcp
Or edit ~/.codex/config.toml (or project .codex/config.toml in a trusted project):
[mcp_servers.odse]
command = "npx"
args = ["-y", "@asobacloud/odse-mcp"]
[mcp_servers.odse.env]
ODSE_WORKSPACE = "/absolute/path/to/your/data"
Codex CLI, the IDE extension, and the ChatGPT desktop Codex host share this config.
Run directly
pip install odse
npx -y @asobacloud/odse-mcp
Tools
| Tool | Runtime | What it does |
|---|---|---|
ListSupportedOEMs |
python3 |
Returns OEM source keys from the installed odse transformer registry, plus odse.__version__. |
ConvertToODSE |
python3 |
Calls odse.transformer.transform on raw CSV/JSON text or a file. Optional source, asset_id, timezone, timeout. If source is omitted, OEM is auto-detected from payload content only. |
ValidateODSERecord |
python3 |
Calls odse.validate_batch on a JSON array of records. |
ConvertToODSE
| Argument | Description |
|---|---|
payload |
Raw telemetry text (CSV or JSON) |
payload_file |
Path to a telemetry file (relative paths resolve against ODSE_WORKSPACE) |
source |
OEM key (e.g. huawei); omit to auto-detect |
asset_id |
Optional asset id forwarded to odse.transform |
timezone |
Optional timezone forwarded to odse.transform (e.g. +02:00) |
timeout |
Timeout in ms (default 120000, max 600000) |
Example (explicit source):
{
"name": "ConvertToODSE",
"arguments": {
"source": "huawei",
"payload": "Time,Active Power(kW),Inverter State\n2024-01-01 12:00:00,12.0,0\n"
}
}
Successful responses are JSON on the tool text channel, for example:
{
"records": [
{
"timestamp": "2024-01-01T12:00:00Z",
"kWh": 1.0,
"error_type": "normal",
"error_code": "0"
}
],
"count": 1,
"source": "huawei",
"odse_version": "0.8.2"
}
Live-fetch arguments (url, headers, body, …) are rejected: not part of this release.
ValidateODSERecord
Pass records as a JSON string of an array of objects. Optional level is forwarded to odse.validate_batch (default schema).
{
"name": "ValidateODSERecord",
"arguments": {
"records": "[{\"timestamp\":\"2024-01-01T12:00:00Z\",\"kWh\":1.0,\"error_type\":\"normal\"}]"
}
}
Validation follows odse.validate_batch package semantics (not a separate JSON Schema checker). Extra properties and numeric bounds behave as the installed odse version defines them.
Tool exit code 1 becomes isError: true on the MCP result. Unexpected crashes are reported as errors.
Requirements
| Runtime | Required for |
|---|---|
| Node.js ≥ 22 | MCP server |
Python 3 + odse |
All tools (pip install odse) |
Architecture
rules/*.json → tool schemas + execution specs
scripts/*.py → odse wrappers (list / convert / validate)
src/loader.ts → bundled rules
src/runner.ts → spawn, timeouts, exit-code → isError
src/index.ts → MCP stdio server
bin/cli.js → npx / bin entrypoint
Development
git clone https://github.com/AsobaCloud/odse-mcp.git
cd odse-mcp
python3 -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install odse
npm install
npm run build
npm test
| Script | Purpose |
|---|---|
npm run build |
Compile TypeScript → dist/ |
npm test |
E2E via real MCP Client + fixture assertions |
npm start |
Run the server on stdio |
npm run dev |
node --watch on dist/ |
Local MCP config (instead of npx):
{
"mcpServers": {
"odse": {
"command": "node",
"args": ["/absolute/path/to/odse-mcp/dist/index.js"],
"env": {
"ODSE_WORKSPACE": "/absolute/path/to/your/data",
"PATH": "/absolute/path/to/odse-mcp/.venv/bin:/usr/bin:/bin"
}
}
}
}
Ensure the PATH (or environment) makes python3 resolve to an interpreter that can import odse.
Releasing
CI runs build + E2E on every push/PR (Node 22 + pip install odse). To publish a new version:
- Bump
versioninpackage.json - Commit, push, and create a GitHub Release (
gh release create vX.Y.Z --generate-notes) .github/workflows/publish.ymlpublishes to npm (Trusted Publisher / OIDC, orNPM_TOKENif configured)
License
MIT © Asoba