AI Agents & MCP

No language model was trained on Bst-Table. Ask a coding agent to "build a Bst-Table grid with batch editing" and it will confidently write some other grid library's code — the very libraries this project exists to replace. The @bloomskill/table-mcp server fixes that: point your agent at it and it gets the real API — searchable docs, the full feature-toggle registry, every cell type, exact type signatures, and the ability to scaffold and validate working grids.

Works with anything that speaks MCP — Claude Code, Cursor, VS Code (Copilot), Claude Desktop.

No MCP? Copy the prompt

Not every assistant speaks MCP, and sometimes you just want to paste something into a chat box. The agent prompt is that: one self-contained briefing — the entry points, the enable* / show* rule, every flag in this release, all the cell types, and the dependencies that are easy to miss — small enough to sit at the top of a conversation.

v0.48.0 · 10 KB · view raw

It is generated from the same corpus as this site and the MCP server, so the flag list is the real one for the version it ships with. Fetch it however suits you:

curl -O https://gitofkumarsathish.github.io/bst-grid/prompt.txt   # the raw file
npx -y @bloomskill/table-mcp prompt                               # print it locally

MCP clients can also read it as the bst://prompt resource, and attach it when handing work to a model that cannot reach the tools. The prompt is the floor; the server is the ceiling — with MCP registered, the agent can look anything up instead of relying on one message.

Register the server

Nothing to install first — npx fetches the package on first run.

Claude Code

claude mcp add bst-table -s user -- npx -y @bloomskill/table-mcp

-s user makes it available in every project on your machine; drop it for the current directory only, or use -s project to write a .mcp.json your team gets on clone.

Cursor · Claude Desktop (mcpServers key)

{
  "mcpServers": {
    "bst-table": {
      "command": "npx",
      "args": ["-y", "@bloomskill/table-mcp"]
    }
  }
}

What the agent gets

The server exposes nine tools:

Tool	What it does
bst_search_docs	full-text search across the corpus
bst_get_feature	a flag's layer, type, default, dependencies and status
bst_get_cell_type	a meta.type cell's value shape, cellMeta and editability
bst_get_api	exact type signatures from the built .d.ts
bst_get_example	runnable usage examples
bst_scaffold_grid	generate a working grid from a feature list
bst_validate_config	check a grid config against the dependency rules
bst_detect_version	resolve which release an API belongs to
bst_list_versions	list the released versions, newest first

Why it stays correct

The corpus is generated from source at build time — the same feature registry, cell types, .d.ts signatures and coverage matrix that produce this documentation site. When a feature ships, it flows into the MCP server automatically, so the agent's knowledge tracks the code release for release. It's the same guarantee these docs give: what you read is what the compiler accepts.

tip

These docs, the agent prompt and the MCP server are three faces of one corpus. Humans read the Feature Guides; agents query bst_get_feature; a chat box gets the prompt. None of them can drift from the code.