Provenance Connect MCP — Technical Reference
Everything you need to connect, authenticate, and query the Provenance Connect MCP connector from Claude or any MCP-compatible client.
What is Provenance Connect?
Provenance Connect is a conversational research connector for SEC filings and corporate disclosure data. It gives Claude access to a curated database of filing signals, classifier-tagged sentences, quarterly fundamentals, and news / press releases — turning natural-language research questions into structured queries against a multi-million-document corpus.
Where most financial data products return filings, Provenance Connect returns signal: language-level patterns computed across the corpus and updated continuously. Users can ask "which companies have rising distress signals" instead of "show me filings with the word bankruptcy."
Universal endpoint:
https://mcp.kscope.io/mcp/ — all users connect to the same URL; identity is carried in the OAuth-issued bearer token.Connection & Authentication
Provenance Connect uses OAuth 2.1 with magic-link sign-in. The flow from discovery to first query takes under two minutes.
Onboarding Flow
1. Find Provenance Connect in the Claude connector directory
(search "Provenance Connect" or "Kaleidoscope")
2. Click Connect → Claude redirects to mcp.kscope.io for OAuth
3. Enter your work email → Provenance Connect checks the allowlist
4. One-time magic link sent to your email → click to authenticate
5. OAuth token issued (30-day JWT) → returned to Claude
6. All subsequent tool calls carry the bearer token automatically
Token Details
| Property | Value |
|---|---|
| Token format | 30-day JWT (Bearer) |
| Access model | Allowlist-gated — only invited email addresses can authenticate |
| Revocation | Takes effect on the next request |
| Client registration | RFC 7591 dynamic client registration |
| PKCE | S256 required |
| Standards | OAuth 2.1, RFC 8414, RFC 9728 |
SSH / CLI users: Device-flow OAuth (RFC 8628) for Claude Code is on the roadmap. In the meantime, authenticate via the browser-based Claude app, which shares the token.
All 9 Tools — Quick Reference
| # | Tool | What it does |
|---|---|---|
| 1 | company_signal_snapshot | Full filing-signal profile for any ticker |
| 2 | cross_company_screening | Find companies matching signal criteria across the universe |
| 3 | peer_comparison | Side-by-side comparison of two or more companies |
| 4 | sector_overview | Sector-level averages, outliers, and trends |
| 5 | quarterly_fundamentals | Revenue, margins, EPS, cash flow from XBRL (up to 20 quarters) |
| 6 | classifier_discovery | Find themed classifiers by keyword across 529 themes |
| 7 | sentence_retrieval | Fetch filing sentences that triggered a classifier for a ticker |
| 8 | semantic_filing_search | Vector-similarity search across ~152M filing sentences |
| 9 | news_search | Vector-similarity search across ~41M news / press-release sentences |
All tools are annotated read-only, non-destructive, and idempotent. Provenance Connect cannot write, modify, or delete any data.
1 — Company Signal Snapshot
company_signal_snapshot
live
Returns the full filing-signal profile for a given ticker: echo trends, composite distress and positive scores, fired classifiers (rising, declining, and statistically anomalous), recent 8-K activity, current price / return data, and latest XBRL fundamentals. This is the primary entry point for single-company research.
"Show me AAPL's current signal profile"
"What are TSLA's top rising classifiers right now?"
Key output fields
| Field | Description |
|---|---|
| echo_composite | Aggregate signal intensity (all classifiers) |
| distress_echo | Aggregate distress-classifier intensity |
| positive_echo | Aggregate positive-signal intensity |
| profile_mutation | How much the company's signal profile has changed vs. prior periods |
| top_rising_classifiers | Classifiers gaining momentum across recent filings |
| notable_z_classifiers | Statistically anomalous classifiers (z-score outliers) |
| eightk_density_90d | Count of 8-K filings in last 90 days |
2 — Cross-Company Screening
cross_company_screening
live
Screens the full universe of ~5,500 tickers against signal criteria. Supports filtering by distress_echo threshold, specific classifier names, SIC sector, and composite score ranges. Returns a ranked list of companies meeting the criteria.
"Which industrials have distress_echo > 10 and profile_mutation > 0.5?"
"Find companies where the going_concern classifier has been rising for 3+ consecutive filings"
3 — Peer Comparison
peer_comparison
live
Side-by-side comparison of two or more companies across filing signals, composite scores, and fundamentals. Useful for relative-value analysis, sector benchmarking, or comparing a company against its direct competitors.
"Compare AAPL, MSFT, and GOOGL on distress_echo and positive_echo"
"How does AMZN's signal profile compare to SHOP right now?"
4 — Sector Overview
sector_overview
live
Provides sector-level (SIC-based) aggregates for key signal metrics — averages, outliers, and directional trends. Useful for macro-level views: which sectors are heating up, which are seeing rising distress, which classifiers are firing most frequently in a given industry.
"Which sectors have the highest average distress_echo this quarter?"
"Show me the top signals in semiconductor companies right now"
5 — Quarterly Fundamentals
quarterly_fundamentals
live
Returns XBRL-parsed financials for a ticker: revenue, gross margin, operating income, net income, EPS, operating cash flow, free cash flow, and balance sheet items — up to 20 quarters. Data is sourced directly from SEC XBRL filings.
"Show NVDA's revenue and margin trend over the last 8 quarters"
"What was AMD's free cash flow trend leading into Q1 2026?"
Coverage note: Gross profit is missing on ~60% of companies due to inconsistent XBRL tagging by filers. A Revenue − COGS fallback is on the roadmap. Coverage is concentrated on ~4,400 tickers.
6 — Classifier Discovery
classifier_discovery
live
Find the right classifier by keyword across 529 themes. Returns matching classifiers with human-readable names, descriptions, directional alpha (positive / negative / neutral), and historical information coefficient. Use this to explore what signals exist before querying them by name.
"What classifiers exist for supply chain themes?"
"Find all distress-related classifiers with positive IC"
Classifier metadata fields
| Field | Description |
|---|---|
| classifier | Internal name (e.g. is_going_concern, cluster_158) |
| human_name | Plain-English label |
| direction | positive = bullish, negative = bearish/distress, neutral = informational |
| llm_weight | LLM confidence weight (1–10); higher = more alpha signal |
| ic | Historical information coefficient vs. forward returns. Typically −0.05 to +0.10 |
7 — Sentence Retrieval
sentence_retrieval
live
Fetches the actual filing sentences that triggered a specific classifier for a given ticker. Returns sentence text, filing date, filing type, accession number, and classifier confidence score. Score threshold: 0.85.
"Show me the exact sentences where PFE's going_concern classifier fired"
"What did TSLA actually say about workforce reductions in their last 10-K?"
8 — Semantic Filing Search
semantic_filing_search
live
Vector-similarity search across ~152M filing sentences using multilingual E5 embeddings (768-dim). Finds content by meaning, not keyword. Supports filtering by company, sector (SIC), form type, and date range.
"Find sentences from 2025 10-Ks about AI infrastructure investment — group by company"
"Which semiconductor firms discussed export controls in late 2024?"
Corpus scope
| Collection | Coverage | Points |
|---|---|---|
| argos_2026 | 2026-01-01 → present | ~5.7M (growing daily) |
| argos_2025 | 2025-01-01 → 2025-12-31 | ~18M |
| argos_2024 | 2024-01-01 → 2024-12-31 | ~17M |
| argos_2022 / 2023 | Partial coverage | ~14M combined |
Forms indexed: 10-K, 10-Q, 8-K, S-1, DEF 14A, 6-K, 20-F, and others. Giant NPORT-P fund filings are intentionally excluded.
9 — News & Press Release Search
news_search
live
Semantic search across ~41M sentences from GlobeNewswire and PRNewswire (2023–2026). Pairs with the filing search tools to triangulate what management files with regulators against what's being announced publicly. This cross-surface pairing is unique to Provenance Connect.
"Find news from Q1 2026 where the issuer's filings also showed rising distress signals"
"Search for FDA approval announcements from biotech companies in 2025"
Data Coverage
| Dataset | Coverage | Count |
|---|---|---|
| SEC filings (with signal aggregates) | All major forms; 2022–2026 | ~262K filings |
| Filing-classifier observations | Score ≥ 0.50 | ~76M observations |
| Filing sentences (vector-searchable) | 2022–2026 (2024–2026 primary) | ~152M sentences |
| Signal classifiers | Distress, growth, capital, governance, sector-specific | 529 classifiers |
| XBRL quarterly fundamentals | Up to 20 quarters per company | ~456K observations, ~4,400 tickers |
| News / press releases | GlobeNewswire + PRNewswire, 2023–2026 | ~41M sentences |
| Company signal snapshots | Current state, refreshed daily | ~5,500 tickers |
Update Cadence
| Collection | Refresh | Latency |
|---|---|---|
| SEC filing sentences | Continuous (sync worker every 5s) | Minutes after SEC acceptance |
| News sentences | Continuous | Minutes after article publish |
| Company signal snapshots | Daily batch | Overnight refresh |
| XBRL fundamentals | As XBRL filings arrive | Within hours of filing |
| Classifier metadata | Weekly | When classifier weights are re-tuned |
Authentication Detail
The Provenance Connect connector implements a full OAuth 2.1 authorization server at https://mcp.kscope.io/.
Endpoints
Authorization: https://mcp.kscope.io/authorize
Token: https://mcp.kscope.io/token
Registration: https://mcp.kscope.io/register # RFC 7591
Metadata: https://mcp.kscope.io/.well-known/oauth-authorization-server
Resource: https://mcp.kscope.io/.well-known/oauth-protected-resource # RFC 9728
Transport
- HTTPS only, TLS 1.2+ via wildcard certificate
- PKCE S256 required on all authorization requests
- Dynamic client registration (no pre-registration step for MCP clients)
Security & Privacy
| Property | Details |
|---|---|
| Authentication | OAuth 2.1 with magic-link sign-in |
| Authorization | Allowlist-gated — only invited email addresses can authenticate |
| Transport | HTTPS only, TLS 1.2+ |
| Access type | Read-only, non-destructive, idempotent — cannot write or delete data |
| PII collected | Email address (allowlist) and OAuth session state only |
| Prompt retention | No prompt content or tool inputs retained beyond standard service logs |
Known Limitations
- No forward-return or backtest performance metrics surfaced through tools
- No index-membership lookups (S&P 500, Russell 1000, etc.)
- No analyst estimates or consensus data
- XBRL fundamentals: gross profit missing on ~60% of companies due to inconsistent XBRL tagging — Revenue − COGS fallback in progress
- Filing sentence coverage: 2024–2026 is primary; 2022–2023 is partial; 2018–2021 not yet loaded
- News ticker tagging occasionally surfaces non-company strings (e.g. exchange names) in ticker fields
- Company signal snapshots are current-state only (no historical time series)
- Device-flow OAuth for SSH / Claude Code users not yet supported
Roadmap
| Item | Status |
|---|---|
| Device-flow OAuth (RFC 8628) for Claude Code / SSH users | Planned |
| Admin CLI for allowlist management | Planned |
| Tiered access (free / pro / enterprise) | Planned |
| XBRL gross-profit fallback (Revenue − COGS) | In Progress |
| Expanded historical sentence coverage (argos_2024 backfill) | In Progress |
| 2018–2021 filing sentence collections | Planned |
Questions, issues, or feature requests? Email raul@kscope.io or use the Request Access form to get in touch.