Skip to content

160-tool MCP server for real-time global intelligence from free public APIs: markets, liquidity, prediction markets, SEC filings, conflict, military, aircraft-cohort early warning, migration and border flows, cyber, climate, news and more. Dated observation history, 29 research prompts, cited briefs, geofences, SSE dashboard. MIT, no paid API keys.

CHANGELOG: state what the branch now does and what stays open

Phoenix Agent (mac-studio)3e77623f2a28
diff --git a/CHANGELOG.md b/CHANGELOG.md
index c0f092d..f2326e7 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,12 +5,37 @@
- Temporal and keyword spike databases now follow `XDG_CACHE_HOME` like
the cache: `$XDG_CACHE_HOME/world-intel-mcp/temporal.db` and
`$XDG_CACHE_HOME/world-intel-mcp/keyword_spikes.db`. With it unset, they
- remain in `~/.cache/world-intel-mcp/`. Users who set `XDG_CACHE_HOME`
- can keep existing baselines by moving these two files from
- `~/.cache/world-intel-mcp/` to `$XDG_CACHE_HOME/world-intel-mcp/`.
- Importing the package no longer opens databases; they open on first use.
+ remain in `~/.cache/world-intel-mcp/`. Existing baselines are not moved:
+ with `XDG_CACHE_HOME` set, the two files already in
+ `~/.cache/world-intel-mcp/` stay there unread and new, empty baselines
+ start under `$XDG_CACHE_HOME/world-intel-mcp/`. To keep their history,
+ move both files there yourself before the next run. Nothing points at
+ the old files at runtime; `intel_weekly_trends` reports the new baseline
+ as created by that call, then as short of history.
+- The cache's default path (`~/.cache/world-intel-mcp/cache.db` when
+ neither `WORLD_INTEL_CACHE_DB` nor `XDG_CACHE_HOME` is set) is resolved
+ when the cache opens, not at import, so it agrees with the baselines if
+ `HOME` changes in between.
+- Importing the package no longer opens databases. The shared cache,
+ circuit breaker, AOI store, vector store and fetcher behind the MCP
+ tools are built on first use, under a lock, as one set: concurrent first
+ uses get the same objects, and a first use that fails binds nothing and
+ closes the databases it had opened, so the next use starts over.
+- `intel_weekly_trends` no longer reports a failed input as zero. When the
+ baseline database is missing or unreadable, `trend_count` is `null` and
+ `error` says why; when anomaly detection fails, `current_anomaly_count`
+ is `null`. A new `data_gaps` list names each failed input, carries
+ anomaly detection's own gaps, and flags a baseline created by the same
+ call or with no metric at the 5 observations volatility needs. Reading
+ the baselines no longer creates an empty database file.
- Apply per-source request spacing to every HTTP retry, including JSON,
text, XML and binary fetches.
+- Known open: each retry now waits out its source's spacing and the
+ breaker is still checked only before the first attempt, so a full outage
+ of a rate-limited source takes longer than before to fall back to its
+ stale copy. The dashboard's Trends panel shows a `null` anomaly count as
+ 0 (it hides the panel only on `error`). The registry tests' temp-cache
+ premise still depends on test order.
## 0.11.0 - 2026-10-12