api-canvas— agent skill

Overview

DataCanvas is a primitive for storage stashes, canvas computes. The existing IStorageProvider is a key/value abstraction — it can stash blobs but exposes no analytical surface. DataCanvas is the analytical surface: register tabular data from upstream APIs, run SQL across multiple registered tables, and export results as CSV/Parquet/JSON.

Tier 3 — @duckdb/node-api is an optional peer dependency (bun add @duckdb/node-api). Servers that don't enable canvas pay zero install cost. Lazy-loaded on first use.

Disabled by default. Set CANVAS_PROVIDER_TYPE=duckdb to enable. Otherwise core.canvas is undefined.

Cloudflare Workers: unsupported. DuckDB has no V8-isolate build. Setting CANVAS_PROVIDER_TYPE=duckdb on a Worker fails closed with a ConfigurationError at init time.

When canvas earns its keep

Two gates before wiring canvas in — both must be yes. Canvas that fails either is a SQL surface nobody queries.

1. Is the data analytical, not just large? Canvas is for tabular/numeric result sets an agent runs SQL over — aggregate, group, join, time-series filter. A discovery/search surface returning categorical metadata (titles, IDs, types, dates) where the workflow is find the record, then drill into it does not qualify, regardless of row count. A 5,000-row search result is still discovery. The gate is shape, not size: the right question is "would an agent write SELECT … GROUP BY against this?", not "does it have many rows?" For name→ID resolution over a bounded list, reach for MCP-side list filtering (see the design-mcp-server skill) instead.
2. Is it too big to inline? A result that fits the response (≤ ~100 rows of compact data) just gets inlined — no canvas. Canvas is the third option only when shape and size both call for it.

If canvas earns its keep, it carries an obligation: a tool that emits a `canvas_id` MUST ship a `dataframe_query` tool in the same server's surface (see the simple-shape Tools row and the Checklist). A canvas_id with no query tool is dead output — the agent literally cannot reach the staged data.

Imports

import type { DataCanvas, CanvasInstance, ColumnSchema } from '@cyanheads/mcp-ts-core/canvas';

The framework wires the optional service onto CoreServices, accessible in the setup() callback — not on `Context`. Handlers access canvas via a module-level accessor:

// src/services/canvas-accessor.ts
import type { DataCanvas } from '@cyanheads/mcp-ts-core/canvas';

let _canvas: DataCanvas | undefined;
export const setCanvas = (c: DataCanvas | undefined) => { _canvas = c; };
export const getCanvas = () => _canvas;

// src/index.ts — wire in setup()
import { setCanvas } from './services/canvas-accessor.js';

await createApp({
  setup(core) {
    setCanvas(core.canvas);
  },
});

interface CoreServices {
  canvas?: DataCanvas;     // present when CANVAS_PROVIDER_TYPE !== 'none'
  // ... other services
}

The token-sharing model

A canvas is identified by an opaque 10-character URL-safe canvasId (~10¹⁸ keyspace). Tools that touch canvas state accept an optional canvas_id input parameter:

When auth is enabled, the effective scope is the composite (tenantId, canvasId). In MCP_AUTH_MODE=none, tenantId collapses to 'default' and the canvasId is the only differentiator — entropy + TTL + the framework's rate limiter make brute-force discovery operationally infeasible. Designed for public-data servers (BrAPI, OpenFEC, etc.). Don't put PII on a no-auth canvas.

Lifecycle

The sweeper runs as an unref'd setInterval — does not keep the event loop alive on its own. Shutdown via core.canvas.shutdown(ctx) (called automatically from ServerHandle.shutdown()) stops the sweeper and tears down every active DuckDB instance.

API

canvas.acquire(maybeId, ctx, options?) → CanvasInstance

Resolves an existing canvas or creates a new one. Returns a {@link CanvasInstance} bound to (canvasId, tenantId). Subsequent operations don't repeat them.

import { getCanvas } from '@/services/canvas-accessor.js';

const canvas = getCanvas();
if (!canvas) throw new Error('DataCanvas is not enabled. Set CANVAS_PROVIDER_TYPE=duckdb.');
const instance = await canvas.acquire(input.canvas_id, ctx);
// instance.canvasId — surface to the agent
// instance.isNew    — true on first call
// instance.expiresAt — ISO 8601 after sliding extension

instance.registerTable(name, rows, options?)

Register an in-memory or async-iterable rowset as a canvas table.

await instance.registerTable('germplasm', rows);

// Explicit schema for AsyncIterable (required — sniffer can't peek).
await instance.registerTable('big_dataset', asyncRows, {
  schema: [
    { name: 'id', type: 'BIGINT' },
    { name: 'label', type: 'VARCHAR', nullable: true },
  ],
});

// Per-table TTL — this table ages on its own clock (30 min sliding window).
// The canvas itself is unaffected; other tables on the same canvas are not touched.
await instance.registerTable('recent_fetch', rows, { ttlMs: 30 * 60 * 1000 });

Schema inference when schema is omitted: sniffer materializes the first 100 rows, unions JS-side types per column, and maps to DuckDB types. All inferred columns are always nullable — a sample can prove a column is nullable, but can never prove NOT NULL (a null may appear past the sniff window). Pass an explicit schema when NOT NULL enforcement is required. Fall-backs to VARCHAR for ambiguous unions (string mixed with numerics). Numeric widening: INTEGER + DOUBLE → DOUBLE, INTEGER + BIGINT → BIGINT. Column ordering follows first-appearance.

Per-table TTL (`ttlMs`) — optional sliding TTL for this table specifically. When set:

• The sweep loop drops the table (and clears its bookkeeping) when its window expires.
• The TTL slides on any read or write against this table: on registerTable (initial set), on query() (both when the table appears in the SQL text and when it is the registerAs target).
• The canvas itself is unaffected — canvas-level expiry is independent.
• Tables registered without ttlMs inherit the canvas lifecycle exactly as before (no change to default behavior).
• instance.describe() surfaces TableInfo.expiresAt (ISO 8601) for tables that have a per-table TTL; absent otherwise.

instance.query(sql, options?)

Run SQL across registered tables. Returns at most rowLimit rows (default 10 000). When the result exceeds rowLimit, the response carries truncated: true and rowCount reflects the number of materialized rows (not the full result set). For full result sets and exact counts, pass registerAs — the result is materialized as a new canvas table; the response carries a preview slice and the exact rowCount.

Querying a table that does not exist throws NotFound (data.reason: 'missing_table') with a recovery hint to re-stage the table or call describe(). This happens when a table has expired (per-table TTL), been dropped, or the name is mistyped. The error is NotFound, not ValidationError — agents should re-stage, not fix the SQL shape.

const result = await instance.query(`
  SELECT germplasmName, COUNT(*) AS n
  FROM germplasm GROUP BY germplasmName ORDER BY n DESC
`);

// Materialize a join result for follow-up queries.
const joined = await instance.query(`
  SELECT g.germplasmName, o.value
  FROM germplasm g JOIN observations o ON g.germplasmDbId = o.germplasmDbId
`, { registerAs: 'g_with_obs', preview: 10 });
// joined.tableName === 'g_with_obs'; joined.rows.length === 10; joined.rowCount === <full count>

// Materialize with a per-table TTL so the chained result ages independently.
const chained = await instance.query(
  'SELECT * FROM recent_fetch WHERE score > 0.8',
  { registerAs: 'high_score', ttlMs: 15 * 60 * 1000 },
);

registerAs rejects with ValidationError (data.reason: 'register_as_clash') if the target name already exists — drop it first.

ttlMs on query({ registerAs }) assigns a per-table TTL to the materialized table — the same sliding semantics as registerTable({ ttlMs }). The SQL text is also scanned for referenced table names; any tracked per-table TTL entry found is slid on each query() call.

denySystemCatalogs?: boolean (default false) — when true, the gate rejects any reference to system catalog namespaces (information_schema, pg_catalog, sqlite_master, duckdb_<name>() calls) at the text-scan layer before the query executes. Use on shared canvases where handle possession is the access boundary — catalog namespaces let callers enumerate every staged handle. Rejection throws ValidationError with data.reason: 'system_catalog_access'. Canvas-token servers that explicitly expose describe() to agents do not need this; only servers that intentionally hide the full catalog should opt in.

Read-only enforcement (four layers + optional catalog layer):

1. Text-level deny-list — pre-parse scan for file/HTTP-reading table functions (read_csv*, read_json*, read_parquet*, read_text, read_blob, glob, iceberg_scan, delta_scan, postgres_scan, mysql_scan, sqlite_scan, plus pre-staged spatial ones).
2. Statement count (must be 1) via extractStatements.
3. Statement type (must be SELECT) via prepared.statementType.
4. EXPLAIN-plan walk against an allowlisted set of physical operators + a denied-function rescan over plan metadata strings.

Any layer's rejection throws ValidationError with a structured data.reason. File-reading scans (READ_CSV, READ_PARQUET, READ_JSON), DDL (CREATE_*, DROP_*, ALTER_*), DML (INSERT, UPDATE, DELETE), exports (COPY_TO_FILE), and utility statements (PRAGMA, ATTACH, LOAD, SET) are all rejected.

instance.registerView(name, selectSql, options?)

Register a SQL view on the canvas. The SELECT runs through the same gate query() enforces (four layers), so a malicious definition fails at registration time, not later when the view is referenced. Pass { denySystemCatalogs: true } to also block catalog namespace references in the view definition — same semantics as the query() flag.

await instance.registerView(
  'sales_by_region',
  'SELECT region, SUM(amount) AS total FROM sales GROUP BY region',
);
// { viewName: 'sales_by_region', columns: ['region', 'total'] }

// Subsequent queries against the view inherit normal gate enforcement at execution time.
const result = await instance.query("SELECT total FROM sales_by_region WHERE region = 'a'");

CREATE OR REPLACE VIEW semantics: re-registering the same name succeeds. Conflict with an existing base table throws validationError({ reason: 'view_table_clash' }).

instance.importFrom(sourceCanvasId, sourceTableName, options?)

Copy a table from another canvas the caller controls into this one. The lifecycle wrapper validates tenancy on both ids before the provider sees either. Round-trips through a sandbox-rooted Parquet temp file so TIMESTAMP/DATE/BLOB columns survive losslessly.

const imported = await target.importFrom(source.canvasId, 'orders', { asName: 'orders_copy' });
// { tableName: 'orders_copy', rowCount: 2, columns: [...] }

Idempotent on re-import (drop + create on the target). asName defaults to sourceTableName. Throws validationError({ reason: 'import_same_canvas' }) if source and target are the same canvas — use query({ registerAs }) to materialize within a single canvas. Throws notFound if the source table is missing; validationError({ reason: 'import_view_clash' }) if the target name collides with an existing view.

instance.export(tableName, target, options?)

Export a canvas table. Path-based exports are sandboxed to CANVAS_EXPORT_PATH (default ./.canvas-exports). Absolute paths and .. traversal are rejected.

// Path target — written inside the sandbox.
await instance.export('g_with_obs', { format: 'parquet', path: 'observations.parquet' });

// Stream target — copied to a temp file in the sandbox, piped to the stream, unlinked.
await instance.export('g_with_obs', { format: 'csv', stream: writableStream });

instance.describe(options?) / instance.drop(name) / instance.clear()

const tables = await instance.describe();
// [{ name: 'germplasm', kind: 'table', rowCount: 200, approxSizeBytes: 8192, columns: [...] }, ...]

// Filter by kind ('table' | 'view').
const onlyViews = await instance.describe({ kind: 'view' });

await instance.drop('staging_table');   // detects kind, emits DROP TABLE or DROP VIEW; false if missing
await instance.clear();                  // returns count dropped (drops views before tables to avoid dependency errors)

TableInfo.kind discriminates 'table' vs 'view'. For views, rowCount is materialized at describe time via COUNT(*) — not free; treat as an approximation if the view is expensive.

TableInfo.approxSizeBytes is set for base tables (DuckDB's estimated_size from duckdb_tables()). It is undefined for views — views have no entry in duckdb_tables(). Use it to decide what to drop when a canvas approaches its memory limit.

Cancellation

registerTable, query, and export accept options.signal: AbortSignal. The provider opens a fresh DuckDB connection per query/export so connection.interrupt() cancels exactly the in-flight work without disturbing other ops on the same canvas.

Result row shape

Rows are returned via DuckDB's getRowObjectsJson() for JSON-safe serialization:

If your tool surfaces row data via structuredContent, the JSON-safe shape flows through unchanged.

Configuration

Minimum viable spillover server

Most canvas use cases are public-data analytics: fetch from an upstream API, stage the full result, let the agent SQL it. The primitives are domain-neutral — canvas.acquire(), spillover(), instance.query() — so the minimum viable shape is small and generic. Reach for it first; add scoping only when a real multi-tenant requirement appears.

Simple-shape defaults

The MCP_HTTP_MAX_BODY_BYTES request-body cap is inbound-only — it bounds the JSON-RPC request, not the upstream data a handler stages into the canvas or the rows it returns. Canvas servers send small requests (queries, SQL, canvas IDs) regardless of dataset size, so the cap never constrains canvas ingestion.

Recipe

A fetcher that spills and a query tool that runs SQL across what was spilled — the whole surface. Swap fetchUpstream for any paginated or streamed source; nothing here is domain-specific.

import { tool, z } from '@cyanheads/mcp-ts-core';
import { spillover } from '@cyanheads/mcp-ts-core/canvas';
import { getCanvas } from '@/services/canvas-accessor.js';

/** Fetch an upstream dataset, inline a preview, spill the full result to a canvas table. */
export const fetchDataset = tool('fetch_dataset', {
  description:
    'Fetch a dataset and stage it on a DataCanvas. Returns an inline preview plus a ' +
    'canvas_id + table you can query with dataframe_query for the full result set.',
  annotations: { readOnlyHint: true },
  input: z.object({
    query: z.string().describe('Upstream search/filter expression'),
    canvas_id: z
      .string()
      .optional()
      .describe('Canvas ID from a prior call. Omit to start fresh — the response returns a new one.'),
  }),
  output: z.object({
    canvas_id: z.string().describe('Canvas ID — pass to dataframe_query or another fetch call'),
    table_name: z.string().describe('Canvas table holding the full result (empty when not spilled)'),
    spilled: z.boolean().describe('True when the result exceeded the preview and was staged'),
    preview: z.array(z.record(z.string(), z.unknown())).describe('Inline rows — the immediate answer'),
    row_count: z.number().describe('Rows staged on the canvas (preview length when not spilled)'),
  }),
  async handler(input, ctx) {
    const canvas = getCanvas();
    if (!canvas) throw new Error('DataCanvas is not enabled. Set CANVAS_PROVIDER_TYPE=duckdb.');

    const instance = await canvas.acquire(input.canvas_id, ctx);
    const result = await spillover({
      canvas: instance,
      source: fetchUpstream(input.query), // any AsyncIterable<Row> | Iterable<Row>
      previewChars: 100_000, // ≈ 25k tokens inline
      signal: ctx.signal,
    });

    return {
      canvas_id: instance.canvasId,
      table_name: result.spilled ? result.handle.tableName : '',
      spilled: result.spilled,
      preview: result.previewRows,
      row_count: result.spilled ? result.handle.rowCount : result.previewRows.length,
    };
  },
});

/** Run read-only SQL across tables staged on a canvas. */
export const dataframeQuery = tool('dataframe_query', {
  description: 'Run a read-only SQL SELECT against tables staged on a canvas by fetch_dataset.',
  annotations: { readOnlyHint: true },
  input: z.object({
    canvas_id: z.string().describe('Canvas ID returned by fetch_dataset'),
    sql: z.string().describe('Read-only SELECT. Reference tables by the names fetch_dataset returned.'),
  }),
  output: z.object({
    rows: z.array(z.record(z.string(), z.unknown())).describe('Result rows (capped at the canvas row limit)'),
    row_count: z.number().describe('Full result count before the row cap'),
  }),
  async handler(input, ctx) {
    const canvas = getCanvas();
    if (!canvas) throw new Error('DataCanvas is not enabled. Set CANVAS_PROVIDER_TYPE=duckdb.');

    const instance = await canvas.acquire(input.canvas_id, ctx);
    const result = await instance.query(input.sql, { signal: ctx.signal });
    return { rows: result.rows, row_count: result.rowCount };
  },
});

When the simple shape is enough

The germplasm-flavored consumer tool template below is the same pattern with domain-specific naming.

Consumer tool template

A domain-specific instance of the minimum viable spillover server above — the same acquire → register → return handle flow with germplasm naming.

import { tool, z } from '@cyanheads/mcp-ts-core';
import { getCanvas } from '@/services/canvas-accessor.js';

export const fetchAndStage = tool('fetch_and_stage_germplasm', {
  description: 'Fetch germplasm matching a query and stage it on a DataCanvas for follow-up SQL.',
  input: z.object({
    query: z.string().describe('Search query'),
    canvas_id: z
      .string()
      .optional()
      .describe(
        'Optional 10-char canvas ID returned from a prior call. Omit on first call to start a fresh canvas; the response will include a new canvas_id you can pass to subsequent calls or share with another agent.',
      ),
  }),
  output: z.object({
    canvas_id: z.string().describe('Canvas ID — pass to subsequent tool calls'),
    is_new_canvas: z.boolean().describe('True if a new canvas was created'),
    table_name: z.string().describe('Canvas table where rows were registered'),
    row_count: z.number().describe('Rows registered'),
    expires_at: z.string().describe('ISO 8601 expiry after sliding 24h window'),
  }),
  async handler(input, ctx) {
    const canvas = getCanvas();
    if (!canvas) {
      throw new Error('DataCanvas is not enabled. Set CANVAS_PROVIDER_TYPE=duckdb.');
    }
    const instance = await canvas.acquire(input.canvas_id, ctx);
    const rows = await fetchGermplasm(input.query);
    const tableInfo = await instance.registerTable('germplasm', rows);
    return {
      canvas_id: instance.canvasId,
      is_new_canvas: instance.isNew,
      table_name: tableInfo.tableName,
      row_count: tableInfo.rowCount,
      expires_at: instance.expiresAt,
    };
  },
});

Pattern: spillover

A handler produces a tabular result that's too big to inline: a paginated REST call that returns 50k rows, a streamed CSV, a database cursor. Inlining everything blows the agent's context; inlining a fixed slice leaves it blind to the rest. Spillover is the third option — show a small preview, register the whole result on the canvas, hand back a token pointing at it. The agent reads the preview directly and reaches for SQL when it needs the rest.

spillover(opts)

import { spillover } from '@cyanheads/mcp-ts-core/canvas';

const result = await spillover({
  canvas: instance,
  source: fetchAllPages(),         // any AsyncIterable<Row> or Iterable<Row>
  previewChars: 100_000,           // ≈ 25k tokens of inline rows
  caps: { maxRows: 50_000 },       // hard upper bound on registered rows
  signal: ctx.signal,
  ttlMs: 30 * 60 * 1000,          // optional: per-table TTL forwarded to registerTable
});

if (result.spilled) {
  // result.previewRows  → inline these in the response
  // result.handle.tableName → surface so the agent can SQL the full set
  // result.truncated    → true if caps.maxRows was hit before the source exhausted
} else {
  // result.previewRows  → entire source fit; no canvas table was created
}

The discriminated union narrows on result.spilled — no runtime checks needed.

Sizing the preview

The budget is characters of `JSON.stringify(row)`, not rows. A row count is a leaky proxy: the same 50 rows is ~500 tokens for compact IDs and ~25k tokens for nested observations. A character budget gives one number that works across heterogeneous tools.

Heuristic: ~4 chars per token for typical JSON. Refine empirically per tool if the row shape is unusual.

Flow

1. Drain. Pull rows, accumulating JSON.stringify(row).length per row, until the running total would exceed previewChars (the row that crosses the budget is the overflow sentinel) or the source exhausts.
2. Source fit. Drain finished under budget — return { spilled: false, previewRows }. No canvas call was made.
3. Source overflows. The sentinel proves there are more rows than fit. Build a merged iterable of (buffered preview rows + sentinel + remaining iterator), hand it to canvas.registerTable, return { spilled: true, previewRows, handle, truncated }.

The merged iterable streams — the helper does not double-buffer the full source.

Schema handling

When the preview budget is small (single-digit rows) and the sniff window matters, pass schema explicitly — the helper's window is only as large as the preview budget allows.

Cancellation and partial state

signal.abort() throws on the next iteration of the preview drain or the spill drain. If abort fires after canvas.registerTable has begun appending rows, the helper best-effort calls canvas.drop(tableName) before the throw propagates — the contract is "partial drain is not registered."

When not to use spillover

• Discovery/search surfaces. A result that's categorical metadata for find-then-drill-in — search hits, ID lookups, catalog browsing — is not analytical and doesn't earn a canvas regardless of row count (see When canvas earns its keep). Use MCP-side list filtering or plain pagination instead.
• Tiny known result. If the upstream call returns ≤ 100 rows, just inline them — no canvas needed.
• Headless register (caller wants the full set on canvas with zero preview rows). Call canvas.registerTable directly. previewChars is rejected at 0; spillover always implies a visible preview.
• Workers runtime. Canvas requires DuckDB native; spillover is a canvas-coupled helper. For Workers parity, persist via ctx.state instead.

Out of scope

• Provenance metadata (source URI, original query). Caller stores externally via ctx.state or tool output — canvas tables carry data only, not lineage.
• Pagination-flavored builder. A paginate(fetchPage) → AsyncIterable<Row> adapter is deferred until a second non-paginated consumer surfaces.
• Token-accurate budget. previewTokens (tokenizer-driven) is a future option; characters cover the common case.
• `caps.maxBytes`. Row caps cover the common case without re-doing serialization the canvas appender skips.

Trade-offs

• DuckDB only in v1. Polars/SQLite/DataFusion don't fit the "agent writes ad-hoc SQL across N registered tables" shape.
• In-memory only. Server restart drops all canvases. For public-data servers, restart is rare and re-fetching upstream data is cheap. Disk persistence is a v2 concern.
• Single process. Tokens issued by one process are not portable to another. Multi-process distributed canvases are out of scope.
• Read-only relative to upstream. Canvas mutations (register, drop, clear, query+registerAs) all stay behind typed methods. Arbitrary SQL cannot mutate.
• No OTel in v1. Canvas operations are not instrumented at the framework level. Add manually via ctx.log if needed.

Platform support

Checklist

• [ ] @duckdb/node-api installed as a peer dependency (bun add @duckdb/node-api)
• [ ] CANVAS_PROVIDER_TYPE=duckdb set in .env
• [ ] Canvas accessor module created (src/services/canvas-accessor.ts or equivalent)
• [ ] Accessor wired in setup() callback via setCanvas(core.canvas)
• [ ] Handler guards for canvas availability (if (!canvas) throw ...)
• [ ] canvas_id accepted as optional input, returned in output
• [ ] A dataframe_query tool is registered in this server whenever any tool emits a canvas_id — a token with no query tool is dead output. Register dataframe_describe too (lets the agent discover staged table/column names)
• [ ] Canvas earns its keep: the staged data is analytical (an agent would SQL it), not a discovery/search surface of categorical metadata
• [ ] SQL queries are read-only (enforced by the four-layer gate, but don't attempt writes)
• [ ] Testing: mock the module-level getCanvas() accessor with vi.spyOn or a test setup that calls setCanvas(mockCanvas)
• [ ] bun run devcheck passes

Related skills

• add-tool — scaffold a new MCP tool definition (use the canvas template above)
• api-config — full env var reference
• api-workers — Worker fail-closed behavior