Boost.space for AI agents

Agents need live operational context to act on — and Boost.space is that layer: your business data aggregated from every tool, unified, and exposed so agents can read it, act on it, and write results back. An agent connected over MCP acts with exactly your Boost.space permissions — it can read, write, and query only the data you can, nothing more.

The MCP is built for production agents, not demos: operations are grouped and pre-computed, so when an agent asks "what's the total invoiced to this client over ten years?" it gets the answer, not the raw history — fewer tokens, and none of the hallucinations that come from dumping raw data into a context window.

And you don't even need to bring a client: the app ships with the Agent Chat (the Agent button in the sidebar) — a built-in agent chat scoped to one module or all of them. Bringing your own agent? Two MCP servers matter - one for your data, one for these docs. Both are listed with their URLs in the MCP overview.

Operate the data over MCP

Connect any MCP client — Claude, ChatGPT, Cursor, Claude Code, and more — using the in-app Connect AI window, which creates the token and shows per-client steps. See Connect via MCP and Common MCP clients.

Once connected, an agent can:

Capability	Example prompt
Search & read records	"Find every company in the Sales space with no email."
Create & update records, deduplicated by key field	"Import these 40 leads — update existing ones by email."
Build structure — modules, fields, field groups, spaces, statuses, labels	"Create a Projects module with owner, status, and due date."
Run Boost.space AI — enrich fields, transform or validate columns	"Standardize the phone-number column in Leads."
Search semantically over the vector database	"Pull the 10 records most relevant to 'unhappy enterprise customers'."
Build & run automations — create them, edit their code, store Secrets, read run logs	"Every Monday at 8:00, email me a summary of new orders."

Read these docs as an agent

Every page of this documentation is machine-readable:

• /llms.txt — an index of all pages with descriptions; /llms-full.txt — the full corpus.
• Any page as markdown — append /index.md to a page URL (e.g. /fields/rollup/index.md) to get the raw, frontmattered source.
• Docs MCP — point an MCP client at https://apidocv5.boost.space/mcp to search and read the docs as tools.
• The Copy Page button on every page copies its markdown for pasting into a prompt.

Why the data model works for agents

• Typed fields — a date is a date, a link resolves to a real record; agents read and write without guessing formats. See Field types.
• Relational links — linked records let an agent traverse from a company to its deals, with lookups and rollups for derived context.
• Safe upserts — the key field means writing the same entity twice updates it instead of duplicating it, and source priority governs conflicts.
• Context where it belongs — space descriptions tell an agent what a dataset holds; the vector database retrieves the few records that matter instead of flooding the context window.
• Permissions built in — tokens are per-user and per-operation; there is no separate security model to configure for agents.

Prefer code?

Everything an agent can do over MCP is also available on the REST API (OpenAPI at /latest.json, markdown rendition at /api-markdown) and the typed SDKs.