Précis Finance MCP¶
Précis Finance MCP lets an AI assistant query your financial model — safely and read-only — over the Model Context Protocol.
You point it at your financial data (in ClickHouse or Databricks), describe your model in a catalogue, and any MCP-capable client — Claude or another agent — can ask questions and get back consistent, defensible numbers: metrics, financial statements, and row-level detail.
What it does¶
- Run metrics and statements — ask for a P&L, a variance, a margin, a utilisation figure, and get a formatted table back.
- Inspect row-level detail — drill from a number into the rows behind it.
- Discover the model — list the scenarios (the datasets you compare: actuals, budgets, forecasts), metrics, and dimensions available before composing a query.
- Charts, exports and Excel — create charts and export JSON/XLSX snapshots, or use live Excel functions.
Numbers come from one place: a catalogue you define on top of a semantic layer — SQL views that state what your data means. The same definition serves every client, so two people asking the same question get the same answer.
What it doesn't do¶
Précis Finance MCP provides read-only financial access. Planning writeback, Python execution and saved report workflows are outside this package. Compatible clients render tables, inspection grids and charts; other clients receive structured results.
It is the open core of the Précis platform, which adds the agentic finance workspace on top of this same engine: a conversational agent and UI, planning with write-back, reports and routines. Nothing here depends on it.
Is this for me?¶
Use Précis Finance MCP if you have financial data you want an AI assistant to query accurately and consistently, on infrastructure you control, without giving it the ability to change anything. You bring the data and a description of your model; it serves the queries.
Where to start¶
| You want to… | Go to |
|---|---|
| Try the hosted demo with synthetic data and no account | Connect to the public demo |
| Run a single-user server in a few minutes | Quickstart |
| Understand the moving parts first | How Précis Finance MCP works |
| Expose a secure multi-user server (bundled Keycloak or your own OIDC IdP) | Remote access — sign-in & identity modes |
| Sign in with your corporate IdP (SSO) through the bundled Keycloak | Keycloak brokering |
| Connect Auth0 / Okta / Entra / Ping directly | External IdP recipes |
| Use live read-only functions in Excel | Précis for Excel |
| Run against bundled or your own ClickHouse (and provision it) | ClickHouse data modes |
| Use Databricks as the analytical backend | Databricks setup |
| Describe your own metrics and statements | Catalogue & semantic model |
| Know what your ClickHouse must contain | ClickHouse schema contract |
| Know what your Databricks warehouse must contain | Databricks schema contract |
| Load your own data | Ingestion & data sources |
| Control what each user may read | User profiles & permissions |
| Wire a new data source, end to end | Onboarding a data source |
| Look up any configuration knob | Environment variable reference |
| See what tools your MCP client gets | MCP tool reference |
| Back it up — and prove it restores | Backups & restore |
| Upgrade to a new release | Upgrading |
| Monitor it and alert on failed loads | Observability |
| Decode an error message | Troubleshooting |
| Answer your security review | Security model |
| Add a metric or dimension to your model | Adding metrics & dimensions |
| Add a custom read tool to the server | Adding read tools |
Stuck, or want a guided evaluation? Help configuring Précis Finance MCP, deployment assistance, and demo environments are an email away — hello@precis.finance.