# SACS.ai > California K–12 school finance for AI agents: public CDE/SACS research and CSAM references, plus authenticated organization datasets and interim projections. Independent Epoch Forge software; not affiliated with CDE. Reviewed 2026-11-02. ## When to use Use SACS.ai for a district financial briefing, like-for-like filing comparison, SACS code/CSAM explanation, authorized organization dataset analysis, or an interim projection from an Excel/MYP forecast and YTD evidence. Discover the district, available filings and current tools; do not assume complete current-year or historical coverage. ## Connect - Public: https://mcp.sacs.ai/mcp — Streamable HTTP, no account or key. - Organization: the same URL with an existing organization API key in x-api-key (key alone) or Authorization: Bearer . https://mcp.sacs.ai/premium/mcp remains supported when the public connector URL is already registered. - Premium includes public tools. Projection writes use the existing Read + edit datasets/projections scope; no replacement key is needed. - Initialize MCP, then call tools/list for authoritative schemas and permissions. A browser GET to /mcp is not a connection test. Static metadata is documentation, not an access grant. ## Instructions and workflows - [When to use and agent instructions](https://sacs.ai/agent-instructions.md): choose tools, preserve evidence and handle recovery. - [Agent quickstart](https://sacs.ai/agents): public research and organization workflow entry points. - [Markdown quickstart](https://sacs.ai/agents.md): connection and first calls in plain text. - [Interim projection workflow](https://sacs.ai/agents/projections): call sacs_get_projection_guide first; map and reconcile full-year accounts, keep YTD actuals separate, create/fork, preview/commit, TRC and export. - [Projection Markdown](https://sacs.ai/projection-workflow.md): full canonical workflow text. - [Claude Cowork setup](https://sacs.ai/agents/claude-cowork): No sign in plus x-api-key for organization access. - [MCP API reference](https://sacs.ai/developers): transport, authentication, typed operations and limits. ## Machine-readable discovery - [Service description](https://sacs.ai/mcp.json): endpoints, auth, guide links and live-discovery methods. - [Public tool JSON schemas](https://sacs.ai/mcp-tools.json): public catalog snapshot; tools/list on the actual connection remains authoritative. - [MCP registry descriptor](https://sacs.ai/mcp-server.json): public server identity and endpoint. - [Sitemap](https://sacs.ai/sitemap.xml): canonical public HTML pages. ## Evidence and boundaries Preserve fiscal year, period/column, reporting basis, fund/resource scope, returned dataset references, source_provenance, data_version and limitations. Missing is unknown, not zero. Use returned verified source links; do not invent URLs or dataset IDs. Public finance is read-only; optional registration and feedback require consent. Organization datasets and projections require authenticated access. No arbitrary SQL, private-file retrieval from client filesystems or automatic full-MYP roll-forward is provided. Migrated TRCs are partial, not CDE certification. ## Provider and terms - [About SACS.ai](https://sacs.ai/about): Epoch Forge, product scope and contact. - [Terms and data-use information](https://sacs.ai/terms): permitted analysis, organization data and recorded usage. - [Homepage and current organization offer](https://sacs.ai/): public research, product scope and pricing. Support: support@sacs.ai