FX Context MCP Server

Find and return individual historical macroeconomic releases whose surprise profiles are most similar to a selected target event. Use this CASE-RETRIEVAL tool when the user wants to identify, rank, inspect, or compare specific historical analog events. It returns the matched events themselves, including event identity, similarity characteristics, surprise profile, and each event's observed post-release reaction. Supported event types: US_CPI and US_NONFARM_PAYROLLS across EURUSD, GBPUSD, and USDJPY. US_PCE is recognized but not yet publicly available: the historical PCE calibration corpus currently has too few usable periods, and a request for US_PCE returns a structured INSUFFICIENT_HISTORICAL_CALIBRATION error (with usable/required event counts) instead of analog results until the corpus grows. ISM is not currently available here: US_ISM_MANUFACTURING and US_ISM_SERVICES have no verified pre-release consensus yet, so they return INSUFFICIENT_HISTORICAL_CALIBRATION rather than analogs. For ISM, use get_event_context for actuals and get_market_reaction for price moves. Similarity methodology is event-specific: US_CPI uses headline/core surprise distance; US_NONFARM_PAYROLLS uses target-relative robust scale normalization (nfp-historical-analog-v1); US_PCE (once activated) uses the same raw surprise -distance approach as US_CPI (pce-historical-analog-v1). Target period: if referencePeriod is omitted, the target is the most recent event of the requested type. Surprises are never estimated, so if that event has no verified pre-release consensus (common right after a new NFP release) the tool fails with NO_VERIFIED_PRE_RELEASE_EXPECTATION. The error details include targetReferencePeriod, latestReleasedPeriod, latestPeriodWithVerifiedExpectation and a hint; to analyze that prior period, retry with referencePeriod=YYYY-MM. Tell the user the newest release could not be analyzed rather than presenting the prior period as the latest. Every "cannot compute" error has the same shape: "CODE: reason [hint] details={json}", where the JSON always contains code, eventType and hint. Do NOT use this tool when the user's primary question is about aggregate behavior across the analog sample; use get_historical_reaction_context instead. The results are historical observations only and do not predict future price direction or provide trading recommendations.

Summarize the aggregate post-event reaction pattern across historical macroeconomic events similar to a selected target release. Use this AGGREGATE-PATTERN tool when the user wants to understand what the historical analog sample collectively showed rather than inspect individual analog events. It summarizes historical outcomes using statistics and behavioral patterns across standardized post-event windows (M5, M15, H1, H4, H24) such as mean, median, quantiles, directional consistency, reaction-path classification (e.g. IMMEDIATE_CONTINUATION, INITIAL_REVERSAL, FADE, DELAYED_REACTION), persistence, and sample-size-based confidence. Supported event types: US_CPI and US_NONFARM_PAYROLLS across EURUSD, GBPUSD, and USDJPY. ISM is not currently available here: US_ISM_MANUFACTURING and US_ISM_SERVICES have no verified pre-release consensus yet, so they return a structured INSUFFICIENT_HISTORICAL_CALIBRATION error rather than aggregate statistics. Similarity methodology is event-specific: US_CPI uses headline/core surprise distance; US_NONFARM_PAYROLLS uses target-relative robust scale normalization (nfp-historical-analog-v1). Target period: if referencePeriod is omitted, the target is the most recent event of the requested type. Surprises are never estimated, so if that event has no verified pre-release consensus (common right after a new NFP release) the tool fails with NO_VERIFIED_PRE_RELEASE_EXPECTATION. The error details include targetReferencePeriod, latestReleasedPeriod, latestPeriodWithVerifiedExpectation and a hint; to analyze that prior period, retry with referencePeriod=YYYY-MM. Tell the user the newest release could not be analyzed rather than presenting the prior period as the latest. Every "cannot compute" error has the same shape: "CODE: reason [hint] details={json}", where the JSON always contains code, eventType and hint. Do NOT use this tool when the user's primary goal is to identify, rank, enumerate, or inspect specific historical analog events. For individual historical cases, use get_historical_analogs. When the user asks for both specific cases and collective pattern analysis, invoke both tools. The results are deterministic empirical observations only and do not predict future prices or provide trading recommendations.

Compare a central bank monetary-policy statement with its immediately preceding comparable statement and return the material changes in policy language, economic assessment, risk balance, and forward guidance. Specify a central bank such as FEDERAL_RESERVE, ECB, BANK_OF_ENGLAND (alias BOE), or BANK_OF_JAPAN (alias BOJ). Optionally specify currentStatementDate in YYYY-MM-DD format to compare a historical statement. If the date is omitted, the latest available statement is compared with the previous one. Returns structured, source-grounded policy changes including direction, materiality, and previous/current evidence. This tool does not provide trading recommendations or predict market movements.

Get the official actual values, pre-release market consensus, and surprise for a specific U.S. macro data release: CPI (US_CPI), Nonfarm Payrolls / jobs report (US_NONFARM_PAYROLLS), PCE inflation (US_PCE), or the ISM Manufacturing / Services PMI (US_ISM_MANUFACTURING / US_ISM_SERVICES). WHEN TO CALL: whenever the user asks what a release printed or how it compared with expectations, even if they do not mention FX Context. Do not answer these from memory; recent releases are not in your training data. Examples: - "What's the latest US PCE?" -> US_PCE - "Was core PCE hotter than expected?" -> US_PCE - "Did CPI beat expectations?" -> US_CPI - "What was the latest NFP?" / "How many jobs were added last month?" -> US_NONFARM_PAYROLLS - "Give me the latest US inflation release." -> US_CPI and/or US_PCE - "What's the latest ISM Manufacturing PMI?" -> US_ISM_MANUFACTURING - "Did ISM Services beat expectations?" -> US_ISM_SERVICES - "What were the actual and consensus values?" / "What was the surprise?" -> the release under discussion Omit referencePeriod for the latest release. For a specific release, pass the month the data covers (YYYY-MM), not the month it was published. WHEN NOT TO CALL: general or educational questions that need no specific release data, e.g. "What is PCE?", "What does NFP stand for?", "Why does CPI matter for FX?", "Explain inflation.", "What is the difference between CPI and PCE?". Use get_upcoming_risks for upcoming release dates, get_market_reaction for price moves after a release, and get_central_bank_changes for central-bank decisions. RESULT: per-metric official actuals (headline/core, MoM/YoY; ISM: headline PMI index) with publisher provenance; when valid, pre-release consensus and a deterministic actual-minus-consensus surprise. Each observation has consensusStatus (AVAILABLE, NOT_FOUND, NOT_CAPTURED, NOT_AVAILABLE, EXCLUDED_POST_RELEASE, NOT_APPLICABLE) explaining any missing consensus, and revisionStatus (REVISED, UNCHANGED, NOT_CAPTURED). Consensus is valid only when capturedAt is before releaseCutoffAt. ISM (US_ISM_MANUFACTURING, US_ISM_SERVICES) currently returns official actuals only: consensusStatus is NOT_CAPTURED and no surprise is computed, so an ISM beat/miss question cannot be answered from this tool yet. Never treat missing consensus or surprise as zero. Official actuals and third-party consensus keep separate source provenance. No trading recommendations or market-direction predictions.

Retrieve deterministic post-event market reaction metrics for an economic release or monetary policy statement. Specify the targetType, such as US_CPI, US_PCE (alias PCE), US_NONFARM_PAYROLLS (alias NFP), US_ISM_MANUFACTURING (alias ISM_MANUFACTURING), US_ISM_SERVICES (alias ISM_SERVICES), FEDERAL_RESERVE, ECB, BANK_OF_ENGLAND (alias BOE), or BANK_OF_JAPAN (alias BOJ). Optionally specify dateOrPeriod: - For US_CPI, US_PCE, US_NONFARM_PAYROLLS, and the ISM PMIs: reference period in YYYY-MM format (e.g. 2024-07). If omitted, the latest event is resolved. - For FEDERAL_RESERVE, ECB, BANK_OF_ENGLAND, or BANK_OF_JAPAN: statement date in YYYY-MM-DD format (e.g. 2024-08-01). If omitted, the latest statement is resolved. Optionally specify instrument: EURUSD, GBPUSD, or USDJPY (defaults to EURUSD). Returns deterministic price action metrics across 5 standardized post-event windows (M5, M15, H1, H4, H24) including startPrice, endPrice, returnPct, high, low, maxUpMovePct, maxDownMovePct, and rangePct. All calculations are strictly deterministic and mathematical. This tool does not provide trading recommendations or predictive signals.

Retrieve known scheduled macroeconomic and central-bank risks that may affect a supported FX instrument within a future time window. Use this tool when the user asks what could move EURUSD, GBPUSD, or USDJPY, what macro risks are coming up, what is happening in the next 24 hours, this week, or over the next month, what upcoming macro risks or scheduled events exist, or when the next CPI, next NFP, next PCE, next ISM, next Fed, next ECB, next BoE, or next BoJ event occurs. The supported risks are U.S. CPI releases, U.S. Nonfarm Payrolls releases, U.S. PCE inflation releases, ISM Manufacturing and Services PMI releases, Federal Reserve monetary-policy decisions, ECB monetary-policy decisions, Bank of England monetary-policy decisions, and Bank of Japan monetary-policy decisions. The horizon can extend up to 720 hours (30 days). scheduleCoverage reports, for each supported event type, the latest known scheduled date and whether the known calendar covers the whole window. If scheduleCoverageComplete is false, some event types may have additional events in the window that are not yet known; say so instead of treating the list as complete. Results are based on official release calendars and contain scheduled events only. This tool does not predict market direction, include unscheduled news, or provide trading recommendations. If no supported scheduled event exists within the requested horizon, an empty risk list is returned.