Précis

Précis Finance MCP · the open core of Précis

A self-hosted FP&A MCP server for governed management reporting.

Précis Finance MCP lets Claude, ChatGPT, and any MCP-capable client answer finance questions from your approved definitions — metrics, hierarchies, statement layouts, and the semantic SQL views behind them. Governed management information for AI clients, read-only by construction and traceable to source. Your own financials, not market data.

View on GitHub Request demo access Book a setup session Read the docs →
  • Self-hosted
  • Docker Compose
  • OAuth 2.1 / dev-key
  • Elastic License 2.0
  • ClickHouse semantic views
  • YAML catalogue
  • Excel add-in
Claude querying governed figures over MCP — sample data.

AI can reach your data. That doesn't mean it understands your numbers.

Point an agent at a generic database connector and it sees raw rows and schemas — not what finance means by revenue, gross margin, utilisation, or revenue by business unit. Two teams define the same KPI two ways. The model infers business meaning that should be explicit. And finance cannot accept an answer it can't trace. A governed read layer closes that gap: one agreed definition per metric, served the same way every time, under access you control.

Compare a database MCP with a governed finance MCP →
Two paths to the same question. Left: an AI agent pointed straight at raw tables and schema guesses at the meaning and returns inconsistent numbers. Right: Précis Finance MCP answers from governed semantic views and a catalogue, returning one agreed definition.

One governed read layer for management information.

Metrics
KPIs defined once, served consistently.
Financial statements
P&L, balance sheet, cash flow, and management layouts from catalogue definitions.
Drill-down
The row-level detail behind a figure, within the views and permissions you expose.
Trends & comparisons
Period, scenario, cost centre, business unit — including measures a generic metric layer can't express.
Catalogue & semantic layer
YAML definitions over plain SQL views, versioned in git; your BI tools can query the same views directly.
Précis Governed result
Project margin · Q1 2026 · €
Cost centre Actual Budget Variance
Advisory · Munich €1,247,800 €1,180,000 +€67,800
Advisory · Berlin €894,200 €920,000 −€25,800
Engineering €2,108,500 €2,050,000 +€58,500
Data & ML €612,400 €680,000 −€67,600
Total €4,862,900 €4,830,000 +€32,900

The same governed figures, in the grid your team already works in.

Finance lives in Excel. The Précis Excel add-in puts the governed read layer one cell away: install a single Office manifest, then call =PRECIS.STATEMENT(…) or =PRECIS.METRIC(…) and the result spills live into the sheet — refreshed on demand, traceable to source, the same definitions every other client sees. =PRECIS.SCENARIOS(), =PRECIS.KPIS() and =PRECIS.HIERARCHY(…) discover what's available; a ribbon button applies Précis block formatting to the spilled range. It's an MCP client of your own instance — it discovers its /mcp endpoint and sign-in from the host that served it, so there's nothing to configure — and it runs in both desktop Excel and Excel on the web.

Explore FP&A reporting in Excel → Function reference
Précis for Excel: the formula =PRECIS.STATEMENT(…) spills a formatted financial statement into the grid — Revenue through EBITDA with Actual, Budget, Budget Variance, full-year and year-on-year columns — alongside the Précis task pane showing a connected, signed-in session and a Refresh button.

Read-only by construction.
Self-hosted by design.

Start with the mechanisms. It runs in your environment, under your own access controls. The server retrieves; it never writes to source. The semantic views are explicit SQL you can read. The catalogue is versioned. No figure is generated by the model: tools return numbers aggregated from source dimensions — account, cost centre, period, scenario — or they say the data isn't there. Identity runs through a local dev key, the bundled Keycloak, or your own OIDC provider (Auth0, Okta, Entra, Ping). You deploy it and operate it under your own security model.

One-directional flow: an AI client calls Précis Finance MCP's read tools, which query the semantic views over your warehouse. There is no write path back to source.

Built for teams putting AI to work on management information.

Finance systems owners

A governed way for AI clients to query finance definitions without exposing raw source systems.

Analytics & data teams

You already maintain warehouse views and metric logic. Précis Finance MCP gives that work an MCP surface finance can trust.

Technical FP&A

You want Claude or ChatGPT answering from your chart of accounts, scenarios, and management reporting layouts.

FP&A consultancies

Configure the finance model once, then support both the open read layer and the full Précis platform.

What ships in Précis Finance MCP — and what doesn't.

Ships (open core, Elastic License 2.0)

  • MCP server & transport
  • Metric engine
  • Financial-statement layouts
  • Semantic SQL view pattern
  • YAML metric catalogue
  • Sample finance model
  • Ingestion path
  • ClickHouse analytical store
  • PostgreSQL platform state
  • Local dev-key mode
  • Multi-user OAuth mode
  • Docker Compose deployment
  • Backup & restore profile
  • Excel add-in (PRECIS.* functions)
  • Configuration & deployment docs

Doesn't ship (lives in Précis)

  • The workspace UI
  • The conversational agent
  • Plan write-back
  • Scenario commit workflows
  • Scheduled Dispatch briefings
  • Report & management-pack builder
  • Excel write-back & round-trip
  • Commercial support (unless agreed)

Open, self-hosted, on your own warehouse.

Most finance MCP options are packaged inside closed SaaS products. Précis Finance MCP is built for teams that want the governed read layer running against their own warehouse — the one you run yourself, with your own definitions. Source-available under Elastic License 2.0: free to use, modify, and self-host commercially; you just can't offer it to others as a hosted service.

Précis Finance MCP runs inside your cloud. An MCP client (Claude, ChatGPT or any MCP-compatible client) connects over HTTPS with OAuth 2.1; only governed answers cross the boundary — never raw records. Inside your perimeter, Précis authenticates via your IdP (Entra, Okta, Auth0 or Keycloak), enforces catalogue, semantic-layer and audit controls, and reads from your BI/warehouse, ERP, EPM and Excel sources.

Running locally in three commands.

export MCP_DEV_KEY=$(openssl rand -hex 32)
docker compose -f deploy/docker-compose.local.yml up -d --build
docker compose -f deploy/docker-compose.local.yml exec precis-mcp \
  python -m precis_mcp.sample_data   # populate the demo model

Then point any MCP client at http://127.0.0.1:8768/mcp with the dev key as bearer token. It proves this isn't a generic SQL MCP — ask: "Show the P&L for FY2025 with comparatives," "Drill revenue down by cost centre," or "Show utilisation by month for the Digital Transformation team."

Full walkthrough in the docs →

The same core, plus the workflow.

The deployment and data model you build here are the foundation the full Précis platform runs on. Précis adds the licensed workspace over the same engine: a conversational agent and UI, planning with user-approved write-back, scenarios, reports and management packs, scheduled briefings, and Excel round-trip. Moving from the open core to Précis is workflow configuration and adoption — not a second data-integration project. Précis prepares; the finance professional decides.

Talk about Précis

Start with the open core.

Request demo access

Bringing IT or finance into the conversation?