mirror of
https://github.com/MODSetter/SurfSense.git
synced 2026-07-09 15:58:30 +00:00
116291a3b6
Rename the import package surfsense_mcp -> mcp_server and remove the src/ layer so the project mirrors the backend's shape (project folder != package name, e.g. surfsense_backend/app). Kills the redundant surfsense_mcp/src/surfsense_mcp nesting. Distribution name and console command (surfsense-mcp) are unchanged; only python -m and internal import paths move to mcp_server. Dockerfile CMD updated; no PYTHONPATH added since the editable install already makes the package importable.
72 lines
2.4 KiB
Python
72 lines
2.4 KiB
Python
"""Shape tool results into the format the caller asked for.
|
|
|
|
Tools default to Markdown (readable for the model and the human watching), and
|
|
can return raw JSON when a caller wants to post-process the data. Large payloads
|
|
are clipped so a single call can't blow the context window.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
from typing import Annotated, Any, Literal
|
|
|
|
from pydantic import Field
|
|
|
|
ResponseFormat = Literal["markdown", "json"]
|
|
|
|
# Shared parameter type for every tool: same name, same semantics everywhere.
|
|
ResponseFormatParam = Annotated[
|
|
ResponseFormat,
|
|
Field(
|
|
description="'markdown' (default, human-readable) or 'json' "
|
|
"(raw data for post-processing)."
|
|
),
|
|
]
|
|
|
|
DEFAULT_CLIP_CHARS = 20_000
|
|
ITEM_FIELD_CLIP_CHARS = 1_500
|
|
|
|
# Fields that duplicate another field verbatim (e.g. Reddit's 'html' mirrors
|
|
# 'body') and only bloat inline results. The full record stays in the run.
|
|
_REDUNDANT_ITEM_FIELDS = frozenset({"html"})
|
|
|
|
|
|
def compact_items(result: Any, field_limit: int = ITEM_FIELD_CLIP_CHARS) -> Any:
|
|
"""Shrink a scraper result for inline return.
|
|
|
|
Drops redundant fields and clips overlong strings per field, so a response
|
|
keeps every item as an excerpt instead of a few items in full. The
|
|
untruncated result remains retrievable via its stored run.
|
|
"""
|
|
if isinstance(result, dict) and isinstance(result.get("items"), list):
|
|
return {
|
|
**result,
|
|
"items": [_compact_item(item, field_limit) for item in result["items"]],
|
|
}
|
|
return result
|
|
|
|
|
|
def _compact_item(item: Any, field_limit: int) -> Any:
|
|
# ponytail: compacts top-level string fields only; nested structures pass
|
|
# through untouched. Upgrade path is a recursive walk if a platform nests
|
|
# long text.
|
|
if not isinstance(item, dict):
|
|
return item
|
|
return {
|
|
key: clip(value, field_limit) if isinstance(value, str) else value
|
|
for key, value in item.items()
|
|
if key not in _REDUNDANT_ITEM_FIELDS
|
|
}
|
|
|
|
|
|
def to_json(payload: Any) -> str:
|
|
"""Pretty-print a payload as JSON, tolerating non-serializable values."""
|
|
return json.dumps(payload, indent=2, ensure_ascii=False, default=str)
|
|
|
|
|
|
def clip(text: str, limit: int = DEFAULT_CLIP_CHARS) -> str:
|
|
"""Trim overlong text, leaving a visible marker of how much was dropped."""
|
|
if len(text) <= limit:
|
|
return text
|
|
dropped = len(text) - limit
|
|
return f"{text[:limit]}\n\n… [{dropped} more characters truncated]"
|