financetoolkit 2.3.3__tar.gz → 2.3.4__tar.gz

{financetoolkit-2.3.3 → financetoolkit-2.3.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: financetoolkit
-Version: 2.3.3
+Version: 2.3.4
 Summary: Transparent and Efficient Financial Analysis
 Project-URL: Homepage, https://www.jeroenbouma.com/projects/financetoolkit
 Project-URL: Documentation, https://www.jeroenbouma.com/projects/financetoolkit/docs
@@ -585,7 +585,7 @@ https://github.com/user-attachments/assets/96ad5288-d83d-4497-a345-1841c48c29d5
 
 ### Remote server
 
-Connect directly to the hosted server at `https://financetoolkit.jeroenbouma.com/mcp`. Nothing needs to be installed locally. On first connection your client opens an OAuth consent page asking for your FMP API key; enter it once and the server handles authentication from there.
+Connect directly to the hosted server at `https://financetoolkit.jeroenbouma.com/mcp`. Nothing needs to be installed locally. On first connection your client opens an OAuth consent page asking for your FMP API key (and, optionally, a FRED API key for the corporate bond series); enter it once and the server handles authentication from there.
 
 | Client | Steps |
 |:---|:---|
@@ -593,8 +593,8 @@ Connect directly to the hosted server at `https://financetoolkit.jeroenbouma.com
 | **Claude.ai** | Customize → Connectors → Add custom connector → paste the URL |
 | **Claude Code** | `claude mcp add --transport http finance-toolkit https://financetoolkit.jeroenbouma.com/mcp` |
 | **VS Code** | Command Palette → MCP: Add Server → HTTP → paste the URL |
-| **Cursor** | Settings → Features → MCP Servers → Add new → http → paste the URL |
-| **Windsurf** | Settings → MCP Servers → Add Server → Remote/HTTP → paste the URL |
+| **Cursor** | Cursor Settings → Tools & MCP → New MCP Server → add `"finance-toolkit": { "url": "<URL>" }` |
+| **Windsurf** | Cascade panel → MCPs → Configure → add `"finance-toolkit": { "serverUrl": "<URL>" }` and refresh |
 
 ### Local installation
 
@@ -776,6 +776,7 @@ The Finance Toolkit has been mentioned in various blogposts, research papers, ne
 - [Fundamental Analysis using Muli-Agent System and Agentic RAG](https://connect.cfauk.org/HigherLogic/System/DownloadDocumentFile.ashx?DocumentFileKey=3ae6f3fc-593a-04ec-c3b8-be5643cab479&forceDialog=0)
 
 **Newsletters & Social Media**
+- [Stack to Street podcast: Your AI Shouldn't Guess the Formula](https://www.stacktostreet.com/episodes/financetoolkit/)
 - [LinkedIn #1](https://www.linkedin.com/posts/jerry-liu-64390071_this-is-an-extremely-comprehensive-100-activity-7340897396471623681-plI4?utm_source=share&utm_medium=member_desktop&rcm=ACoAABhYAQUBkaNnlBjz_IYBRfaurU68rfoZPE8) and [Linkedin #2](https://www.linkedin.com/posts/pyquant-news_getting-clean-fundamental-data-used-to-be-activity-7314273012927651841-KXP3?utm_source=share&utm_medium=member_desktop&rcm=ACoAABhYAQUBkaNnlBjz_IYBRfaurU68rfoZPE8)
 - [Facebook](https://www.facebook.com/groups/techtitansgroup/posts/1515790739748239/)
 - [Quant Seeker](https://www.quantseeker.com/p/weekly-research-recap)

{financetoolkit-2.3.3 → financetoolkit-2.3.4}/README.md

@@ -540,7 +540,7 @@ https://github.com/user-attachments/assets/96ad5288-d83d-4497-a345-1841c48c29d5
 
 ### Remote server
 
-Connect directly to the hosted server at `https://financetoolkit.jeroenbouma.com/mcp`. Nothing needs to be installed locally. On first connection your client opens an OAuth consent page asking for your FMP API key; enter it once and the server handles authentication from there.
+Connect directly to the hosted server at `https://financetoolkit.jeroenbouma.com/mcp`. Nothing needs to be installed locally. On first connection your client opens an OAuth consent page asking for your FMP API key (and, optionally, a FRED API key for the corporate bond series); enter it once and the server handles authentication from there.
 
 | Client | Steps |
 |:---|:---|
@@ -548,8 +548,8 @@ Connect directly to the hosted server at `https://financetoolkit.jeroenbouma.com
 | **Claude.ai** | Customize → Connectors → Add custom connector → paste the URL |
 | **Claude Code** | `claude mcp add --transport http finance-toolkit https://financetoolkit.jeroenbouma.com/mcp` |
 | **VS Code** | Command Palette → MCP: Add Server → HTTP → paste the URL |
-| **Cursor** | Settings → Features → MCP Servers → Add new → http → paste the URL |
-| **Windsurf** | Settings → MCP Servers → Add Server → Remote/HTTP → paste the URL |
+| **Cursor** | Cursor Settings → Tools & MCP → New MCP Server → add `"finance-toolkit": { "url": "<URL>" }` |
+| **Windsurf** | Cascade panel → MCPs → Configure → add `"finance-toolkit": { "serverUrl": "<URL>" }` and refresh |
 
 ### Local installation
 
@@ -731,6 +731,7 @@ The Finance Toolkit has been mentioned in various blogposts, research papers, ne
 - [Fundamental Analysis using Muli-Agent System and Agentic RAG](https://connect.cfauk.org/HigherLogic/System/DownloadDocumentFile.ashx?DocumentFileKey=3ae6f3fc-593a-04ec-c3b8-be5643cab479&forceDialog=0)
 
 **Newsletters & Social Media**
+- [Stack to Street podcast: Your AI Shouldn't Guess the Formula](https://www.stacktostreet.com/episodes/financetoolkit/)
 - [LinkedIn #1](https://www.linkedin.com/posts/jerry-liu-64390071_this-is-an-extremely-comprehensive-100-activity-7340897396471623681-plI4?utm_source=share&utm_medium=member_desktop&rcm=ACoAABhYAQUBkaNnlBjz_IYBRfaurU68rfoZPE8) and [Linkedin #2](https://www.linkedin.com/posts/pyquant-news_getting-clean-fundamental-data-used-to-be-activity-7314273012927651841-KXP3?utm_source=share&utm_medium=member_desktop&rcm=ACoAABhYAQUBkaNnlBjz_IYBRfaurU68rfoZPE8)
 - [Facebook](https://www.facebook.com/groups/techtitansgroup/posts/1515790739748239/)
 - [Quant Seeker](https://www.quantseeker.com/p/weekly-research-recap)

{financetoolkit-2.3.3 → financetoolkit-2.3.4}/financetoolkit/cache/cache_controller.py

@@ -22,7 +22,7 @@ from financetoolkit.cache import (
     serialization_model,
 )
 from financetoolkit.cache.sqlite_model import SCHEMA_VERSION, SQLiteBackend
-from financetoolkit.utilities import logger_model
+from financetoolkit.utilities import logger_model, sources_model
 
 logger = logger_model.get_logger()
 
@@ -433,8 +433,8 @@ class Cache:
             parameters (dict[str, Any] | None): Parameters that change the data.
             date_axis (int): 0 when the index holds the dates, 1 when the columns do. Defaults to 0.
             reference_date (date | None): The date the revision window is measured
-                back from when the requested range has no end or ends after it. Defaults to today;
-                mainly useful for testing.
+                back from when the requested range has no end or ends after it. Defaults to None,
+                which uses today; mainly useful for testing.
 
         Returns:
             CachePlan: The cached data per entity and the gaps still to be fetched.
@@ -529,6 +529,9 @@ class Cache:
                     with contextlib.suppress(Exception):
                         self._backend.delete_entity(key, entity)
 
+        if plan.cached:
+            sources_model.record_source(source)
+
         return plan
 
     def store(
@@ -682,7 +685,12 @@ class Cache:
             if time.time() - updated_at > effective_ttl:
                 return None
 
-            return serialization_model.decode_object(payload)
+            value = serialization_model.decode_object(payload)
+            # A cached MCP answer names the sources it was computed from itself.
+            if source != policy_model.MCP:
+                sources_model.record_source(source)
+
+            return value
         except Exception as error:
             # Any read failure is reported as a miss, so the caller falls back to the source.
             logger.debug(

{financetoolkit-2.3.3 → financetoolkit-2.3.4}/financetoolkit/cache/frame_model.py

@@ -124,6 +124,10 @@ def slice_frame(
     the caller. Labels that cannot be parsed as dates are kept, which matters for
     statement frames that carry non-date columns alongside the reporting periods.
 
+    A period is kept when it overlaps the window, as a fresh request truncates it: a
+    start date of 31 December 2014 keeps the year 2014. A cached frame is then the same
+    as the one a fresh request returns, so no answer depends on what the cache holds.
+
     Args:
         data (pd.DataFrame | pd.Series): The frame to slice.
         start (str | date | None): Inclusive start of the window. None leaves it open.
@@ -147,9 +151,14 @@ def slice_frame(
         # is not a date.
         keep = vectorised.isna()
         within = np.ones(len(axis), dtype=bool)
+        last_days = (
+            axis.to_timestamp(how="end").normalize()
+            if isinstance(axis, pd.PeriodIndex)
+            else vectorised
+        )
 
         if start_date is not None:
-            within &= vectorised >= pd.Timestamp(start_date)
+            within &= last_days >= pd.Timestamp(start_date)
         if end_date is not None:
             within &= vectorised <= pd.Timestamp(end_date)
 
@@ -167,7 +176,8 @@ def slice_frame(
             mask.append(True)
             continue
 
-        within_start = start_date is None or label_date >= start_date
+        last_day = label.end_time.date() if isinstance(label, pd.Period) else label_date
+        within_start = start_date is None or last_day >= start_date
         within_end = end_date is None or label_date <= end_date
         mask.append(within_start and within_end)
 

{financetoolkit-2.3.3 → financetoolkit-2.3.4}/financetoolkit/cache/policy_model.py

@@ -41,6 +41,8 @@ NORGES_BANK = "NorgesBank"
 EIOPA = "EIOPA"
 INTERNATIONAL_MONETARY_FUND = "IMF"
 SHILLER = "Shiller"
+WORLD_BANK = "WorldBank"
+IMF_PORTWATCH = "IMFPortWatch"
 CBOE = "Cboe"
 STOXX = "STOXX"
 RESERVE_BANK_OF_AUSTRALIA = "ReserveBankOfAustralia"
@@ -110,8 +112,8 @@ def get_last_closed_session(now: datetime | None = None) -> date:
     date.
 
     Args:
-        now (datetime | None): The moment to judge, with a time zone. Defaults to the
-            current time; mainly useful for testing.
+        now (datetime | None): The moment to judge, with a time zone. Defaults to None,
+            which uses the current time; mainly useful for testing.
 
     Returns:
         date: The last date whose daily bars are final.
@@ -244,6 +246,7 @@ POLICIES: dict[str, CachePolicy] = {
     f"{BANK_OF_ENGLAND}.millennium": CachePolicy(ttl_seconds=365 * DAY),
     f"{BANK_OF_ENGLAND}.curve_current_month": CachePolicy(ttl_seconds=DAY),
     f"{BANK_OF_JAPAN}.series": CachePolicy(ttl_seconds=DAY, revision_days=31),
+    f"{BANK_OF_JAPAN}.price_index": CachePolicy(ttl_seconds=DAY),
     f"{EUROPEAN_CENTRAL_BANK}.economics_series": CachePolicy(
         ttl_seconds=DAY, revision_days=31
     ),
@@ -252,6 +255,9 @@ POLICIES: dict[str, CachePolicy] = {
     f"{STATISTICS_BUREAU_OF_JAPAN}.series": CachePolicy(ttl_seconds=DAY),
     # The keyless BLS API allows 25 requests a day, so the latest releases are fetched once.
     f"{BUREAU_OF_LABOR_STATISTICS}.series": CachePolicy(ttl_seconds=DAY),
+    # The consumer prices are one request for every measure, kept for six hours so the
+    # month a release publishes appears the same day.
+    f"{BUREAU_OF_LABOR_STATISTICS}.consumer_prices": CachePolicy(ttl_seconds=DAY // 4),
     # A past year of the Treasury yield curve is final; only the current year changes.
     f"{US_TREASURY}.par_yield_curve": CachePolicy(ttl_seconds=DAY),
     f"{US_TREASURY}.par_yield_curve_year": CachePolicy(ttl_seconds=30 * DAY),
@@ -287,6 +293,11 @@ POLICIES: dict[str, CachePolicy] = {
     f"{INTERNATIONAL_MONETARY_FUND}.codelist": CachePolicy(ttl_seconds=30 * DAY),
     # Updated about once a month.
     f"{SHILLER}.stock_market_data": CachePolicy(ttl_seconds=7 * DAY),
+    # The Pink Sheet is updated in the first days of every month.
+    f"{WORLD_BANK}.commodity_prices": CachePolicy(ttl_seconds=7 * DAY),
+    # PortWatch adds the latest days once a week, on Tuesdays, and the chokepoints rarely.
+    f"{IMF_PORTWATCH}.chokepoint_transits": CachePolicy(ttl_seconds=DAY),
+    f"{IMF_PORTWATCH}.chokepoints": CachePolicy(ttl_seconds=30 * DAY),
     # Index histories published whole, one file per index, updated after every close.
     f"{CBOE}.index": CachePolicy(ttl_seconds=DAY),
     f"{CBOE}.option_chain": CachePolicy(ttl_seconds=900),
@@ -316,6 +327,11 @@ POLICIES: dict[str, CachePolicy] = {
     f"{INTERNATIONAL_MONETARY_FUND}.consumer_prices": CachePolicy(
         ttl_seconds=DAY, revision_days=365
     ),
+    # Government finance and national accounts, published quarterly or yearly and revised
+    # back years when a country revises its national accounts.
+    f"{INTERNATIONAL_MONETARY_FUND}.government_finance": CachePolicy(
+        ttl_seconds=7 * DAY, revision_days=1095
+    ),
     f"{EUROPEAN_CENTRAL_BANK}.convergence_yields": CachePolicy(
         ttl_seconds=DAY, revision_days=93
     ),