yaml-flow 8.4.25 → 8.5.1

package/examples/board/server/README-mcp-api.md

@@ -0,0 +1,680 @@
+# Board Server — MCP API Reference
+
+Every board server exposes two MCP endpoints under `/api/boards/:boardId`:
+
+| Endpoint | Content-Type | Purpose |
+|---|---|---|
+| `POST /api/boards/:boardId/mcp` | `application/json` → `application/json` | All tools except file downloads |
+| `POST /api/boards/:boardId/mcp-raw` | `application/json` → `application/octet-stream` | File content download only (`inspect.file-contents`) |
+
+## Request / response shape
+
+```json
+// POST /api/boards/:boardId/mcp
+{
+  "tool": "<tool-name>",
+  "args": { ... }
+}
+
+// Response (200)
+{
+  "status": "success",
+  "data": { ... }
+}
+
+// Error (400 / 500)
+{
+  "error": "message"
+}
+```
+
+---
+
+## Tool reference
+
+### `discover.source-kinds`
+
+Returns the known source-kind registry from the task executor.
+
+**Args:** none
+
+**Returns:**
+```json
+{
+  "status": "success",
+  "data": {
+    "version": "...",
+    "commonSourceFields": { ... },
+    "sourceKinds": { "<kind>": { ... } }
+  }
+}
+```
+
+---
+
+### `inspect.board-runtime-status`
+
+Returns a summary of all registered cards and overall board state.
+
+**Args:** none
+
+**Returns:**
+```json
+{
+  "status": "success",
+  "data": {
+    "meta": { ... },
+    "summary": {
+      "card_count": 2,
+      "completed": 1,
+      "eligible": 0,
+      "pending": 0,
+      "blocked": 0,
+      "in_progress": 1,
+      "failed": 0,
+      "unresolved": 0
+    },
+    "cards": [
+      {
+        "card-id": "my-card",
+        "status": "completed",
+        "error": null,
+        "requires": [],
+        "requires_satisfied": [],
+        "requires_missing": [],
+        "provides_declared": ["output-key"],
+        "provides_runtime": ["output-key"]
+      }
+    ]
+  }
+}
+```
+
+---
+
+### `inspect.card-definition-and-runtime`
+
+Returns the full card definition, its runtime state, and the current values of all `provides` outputs.
+
+**Args:**
+
+| Field | Type | Required |
+|---|---|---|
+| `card_id` | string | yes |
+
+**Returns:**
+```json
+{
+  "status": "success",
+  "data": {
+    "cardId": "my-card",
+    "card_status_in_board": {
+      "name": "my-card",
+      "status": "completed",
+      "error": null,
+      "requires": [],
+      "requires_satisfied": [],
+      "requires_missing": [],
+      "provides_declared": ["output-key"],
+      "provides_runtime": ["output-key"]
+    },
+    "card_definition_and_static_data": {
+      "id": "my-card",
+      "card_data": { ... },
+      "source_defs": [ ... ]
+    },
+    "refs_for_fetched_source_files": {
+      "output.json": "sha256:abc123..."
+    },
+    "runtime_data": {
+      "requires": { "upstream-key": { ... } },
+      "provides": { "output-key": { ... } },
+      "computed_values": { ... },
+      "rendered_view": {
+        "layout": "...",
+        "features": { ... },
+        "elements": [ { "id": "el1", "kind": "text", "label": "Output", "visible": true, "resolved": "..." } ]
+      }
+    }
+  }
+}
+```
+
+---
+
+### `inspect.chat-messages-on-cards`
+
+Returns chat records for one or all cards, with flexible filtering.
+
+**Args:**
+
+| Field | Type | Default | Notes |
+|---|---|---|---|
+| `card_id` | string | — | Target card |
+| `all-turns` | boolean | `false` | Return all turns (ignores `tail-turns`) |
+| `tail-turns` | number | — | Last N user turns only |
+| `tail` | number | — | Last N individual messages |
+| `turn-id` | string | — | Only messages with this turn id |
+| `tail-turns-before-id` | string | — | Requires `tail-turns`; messages before the given turn id |
+
+**Returns:**
+```json
+{
+  "status": "success",
+  "data": {
+    "cardId": "my-card",
+    "messages": [
+      { "role": "user",      "text": "hello",     "turn": "abc123", "ts": "2026-05-28T10:00:00.000Z" },
+      { "role": "system",   "text": "processing...", "turn": "abc123", "ts": "2026-05-28T10:00:00.100Z" },
+      { "role": "assistant","text": "hi there",  "turn": "abc123", "ts": "2026-05-28T10:00:01.200Z" }
+    ]
+  }
+}
+```
+
+When a system message references a file attachment, a `retrieval_hint` field is added:
+```json
+{ "role": "system", "text": "file uploaded: report.pdf as abc.pdf #0", "turn": "abc123",
+  "retrieval_hint": "Retrieve using inspect-file-contents --card-id my-card --file-idx 0" }
+```
+
+---
+
+### `inspect.file-contents` _(only on `/mcp-raw`)_
+
+Downloads the raw bytes of a file attachment stored on a card.
+
+**Endpoint:** `POST /api/boards/:boardId/mcp-raw`
+
+**Args:**
+
+| Field | Type | Required |
+|---|---|---|
+| `card_id` | string | yes |
+| `file_idx` | number | yes |
+
+**On `/mcp-raw`:** returns raw bytes (`application/octet-stream`) with headers:
+- `Content-Disposition: attachment; filename="<name>"`
+- `Content-Type: <mime_type>`
+
+**On `/mcp`:** this tool is rejected. Use `/mcp-raw`.
+
+---
+
+### `manage.read-card`
+
+Reads the stored document for a single live card.
+
+**Args:**
+
+| Field | Type | Required |
+|---|---|---|
+| `card_id` | string | yes |
+
+**Returns:** array containing the matching live card document (empty array if not found), wrapped in the MCP success envelope:
+```json
+{
+  "status": "success",
+  "data": [
+    {
+      "id": "my-card",
+      "card_data": {
+        "id": "my-card",
+        "sources": [ ... ],
+        "compute": { ... },
+        "files": [ { "name": "report.pdf", "stored_name": "abc.pdf", "mime_type": "application/pdf", "size": 1024 } ]
+      }
+    }
+  ]
+}
+```
+
+---
+
+### `manage.upsert-card`
+
+Validates, stores, and registers a card definition. Triggers a board restart for the card.
+
+**Args:**
+
+| Field | Type | Required |
+|---|---|---|
+| `card_id` | string | yes |
+| `candidate_card_content` | object | yes — must include `id` matching `card_id` |
+
+**Returns (success):**
+```json
+{
+  "status": "success",
+  "data": {
+    "validation": { "status": "success", "data": { "isValid": true } },
+    "card_saved": null,
+    "board_result": { "status": "success" },
+    "refresh_notify": { "status": "success" }
+  }
+}
+```
+
+**Returns (validation failure):**
+```json
+{ "error": "Validation failed" }
+```
+
+---
+
+### `manage.deprecate`
+
+Removes a card from the board.
+
+**Args:**
+
+| Field | Type | Required |
+|---|---|---|
+| `card_id` | string | yes |
+
+**Returns:** the `board.removeCard` result normalized into the MCP success envelope. On success:
+```json
+{ "status": "success", "data": {} }
+```
+
+---
+
+### `manage.upload-card-file`
+
+Uploads a file to a card's attachment store (outside the chat flow).
+
+**Args — provide exactly one of `bytes`, `text`, or `base64`:**
+
+| Field | Type | Notes |
+|---|---|---|
+| `card_id` | string | required |
+| `file_name` | string | required |
+| `content_type` | string | optional, default `application/octet-stream` |
+| `bytes` | number[] | raw byte array |
+| `text` | string | UTF-8 encoded |
+| `base64` | string | base64 or base64url encoded |
+
+**Returns:** the stored file metadata object in the MCP success envelope:
+```json
+{
+  "status": "success",
+  "data": {
+    "ok": true,
+    "file": {
+      "fileIdx": 2,
+      "name": "report.pdf",
+      "stored_name": "abc123.pdf",
+      "mime_type": "application/pdf",
+      "size": 204800,
+      "uploaded_at": "2026-05-28T10:00:00.000Z"
+    }
+  }
+}
+```
+
+---
+
+### `stage-ai-response-and-any-attachments`
+
+Stages an assistant response (with optional file attachments) directly into a card's chat store. Used by agent pipelines to inject a response without going through the SSE chat flow.
+
+**Args:**
+
+| Field | Type | Notes |
+|---|---|---|
+| `card_id` | string | required |
+| `text` | string | response text |
+| `turn-id` | string | turn id to associate the message with |
+| `files` | array | optional array of file metadata objects |
+
+**Returns:**
+```json
+{
+  "status": "success",
+  "data": {
+    "cardId": "my-card",
+    "id": "msg-uuid-...",
+    "role": "assistant",
+    "turn": "abc123",
+    "files": []
+  }
+}
+```
+
+---
+
+## `preflight.*` tools
+
+Preflight tools let you test a **candidate card definition** in-memory without committing it to the board.
+
+### Quick Reference
+
+All preflight tools are invoked through:
+
+```json
+{
+  "tool": "preflight.<name>",
+  "args": { ... }
+}
+```
+
+| Tool | MCP `args` shape | Success `data` shape |
+|---|---|---|
+| `preflight.validate-candidate-card-definition` | `{ "candidate_card_content": <card> }` | `{ "cardId": string, "isValid": boolean, "issues": string[] }` |
+| `preflight.materialize-candidate-card` | `{ "candidate_card_content": <card>, "mock_requires": {}, "mock_fetched_sources": {} }` | `{ "cardId": string, "ok": boolean, "computed_values": object, "errors": Array<{ bindTo, error }>, "provides_outputs": object, "rendered_view": object }` |
+| `preflight.probe-single-source-in-candidate-card` | `{ "candidate_card_content": <card>, "source_idx": number, "mock_projections": {} }` | `{ "bindTo": string, "reachable": boolean, "latencyMs"?: number, "note"?: string }` |
+| `preflight.run-single-source-in-candidate-card` | `{ "candidate_card_content": <card>, "source_idx": number, "mock_projections": {} }` | `{ "bindTo": string, "ok": boolean, "result": unknown, "issues": string[] }` |
+| `preflight.run-single-source-in-live-card` | `{ "card_id": string, "source_idx": number, "mock_requires": {} }` | `{ "bindTo": string, "ok": boolean, "result": unknown, "issues": string[] }` |
+| `preflight.run-one-cycle-with-candidate-card` | `{ "candidate_card_content": <card>, "mock_requires": {} }` | `{ "cardId": string, "ok": boolean, "issues": string[], "provides_outputs": object, "rendered_view": object }` |
+
+Unless otherwise noted, successful HTTP responses use the normal MCP envelope:
+
+```json
+{
+  "status": "success",
+  "data": { ... }
+}
+```
+
+### `preflight.validate-candidate-card-definition`
+
+Schema-validates a candidate card.
+
+**Args:**
+
+| Field | Type | Required |
+|---|---|---|
+| `candidate_card_content` | object | yes |
+
+**MCP request body:**
+```json
+{
+  "tool": "preflight.validate-candidate-card-definition",
+  "args": {
+    "candidate_card_content": { "id": "my-card", "card_data": {} }
+  }
+}
+```
+
+**Returns (valid):**
+```json
+{ "status": "success", "data": { "cardId": "my-card", "isValid": true, "issues": [] } }
+```
+
+**Returns (invalid):** outer `status` is still `"success"` — the validator ran cleanly, it just found problems:
+```json
+{ "status": "success", "data": { "cardId": "my-card", "isValid": false, "issues": [ "view.elements[0].kind is required", "source_defs[0].bindTo is duplicate" ] } }
+```
+
+---
+
+### `preflight.materialize-candidate-card`
+
+Evaluates the card's `compute[]` steps with the supplied mocks. Does **not** run sources and does **not** produce a rendered view — use `preflight.run-one-cycle-with-candidate-card` for a full pipeline simulation.
+
+**Args:**
+
+| Field | Type | Notes |
+|---|---|---|
+| `candidate_card_content` | object | required |
+| `mock_requires` | object | required — may be empty, but the key must be present |
+| `mock_fetched_sources` | object | required — may be empty, but the key must be present |
+
+**MCP request body:**
+```json
+{
+  "tool": "preflight.materialize-candidate-card",
+  "args": {
+    "candidate_card_content": { "id": "my-card", "card_data": {}, "compute": [] },
+    "mock_requires": {},
+    "mock_fetched_sources": {}
+  }
+}
+```
+
+**Returns:**
+```json
+{
+  "status": "success",
+  "data": {
+    "cardId": "my-card",
+    "ok": true,
+    "computed_values": { "title": "My Report", "count": 42 },
+    "errors": [],
+    "provides_outputs": {
+      "output-key": { "title": "My Report", "count": 42 }
+    },
+    "rendered_view": {
+      "layout": "stack",
+      "features": {},
+      "elements": [
+        { "id": "summary", "kind": "text", "label": "Summary", "visible": true, "resolved": "My Report" }
+      ]
+    }
+  }
+}
+```
+
+`errors` is an array of `{ "bindTo": string, "error": string }`. On compute failure `ok` is `false` and `errors` lists each broken binding:
+```json
+{ "status": "success", "data": { "cardId": "my-card", "ok": false, "computed_values": {}, "errors": [ { "bindTo": "title", "error": "undefined variable" } ], "provides_outputs": {}, "rendered_view": { "layout": null, "features": null, "elements": [] } } }
+```
+
+If `mock_requires` or `mock_fetched_sources` is omitted entirely, the MCP call fails with a 400-style error such as `{ "error": "MCP tool requires mock_requires" }`.
+
+---
+
+### `preflight.probe-single-source-in-candidate-card`
+
+Calls the registered task executor's `probe-source-preflight` hook for a single source in a candidate card — tests reachability without fetching data. **Requires a task executor to be registered**; returns an error if none is configured.
+
+**Args:**
+
+| Field | Type | Notes |
+|---|---|---|
+| `candidate_card_content` | object | required |
+| `source_idx` | number | index into the card's `source_defs` array |
+| `mock_projections` | object | optional mock projection data |
+
+**MCP request body:**
+```json
+{
+  "tool": "preflight.probe-single-source-in-candidate-card",
+  "args": {
+    "candidate_card_content": { "id": "my-card", "card_data": {}, "source_defs": [ ... ] },
+    "source_idx": 0,
+    "mock_projections": {}
+  }
+}
+```
+
+**Returns:** connectivity / readiness metadata for the source — does **not** return fetched data:
+```json
+{
+  "status": "success",
+  "data": {
+    "bindTo": "my-source",
+    "reachable": true,
+    "latencyMs": 45,
+    "note": "endpoint reachable"
+  }
+}
+```
+
+> If no task executor is configured the call fails with an HTTP error such as `{ "error": "No task-executor registered for this board" }`.
+
+---
+
+### `preflight.run-single-source-in-candidate-card`
+
+Executes a single source in a candidate card and returns the real fetched output. This path is live-fetch-only and **requires a configured task executor**.
+
+**Args:**
+
+| Field | Type | Notes |
+|---|---|---|
+| `candidate_card_content` | object | required |
+| `source_idx` | number | index into the card's `source_defs` array |
+| `mock_projections` | object | optional mock projection data |
+
+**MCP request body:**
+```json
+{
+  "tool": "preflight.run-single-source-in-candidate-card",
+  "args": {
+    "candidate_card_content": { "id": "my-card", "card_data": {}, "source_defs": [ ... ] },
+    "source_idx": 0,
+    "mock_projections": {}
+  }
+}
+```
+
+**Returns:**
+```json
+{
+  "status": "success",
+  "data": {
+    "bindTo": "my-source",
+    "ok": true,
+    "result": { "items": [ ... ] },
+    "issues": []
+  }
+}
+```
+
+On live fetch failure the command still returns `status: "success"`, but with `ok: false` and a populated `issues` list:
+
+```json
+{
+  "status": "success",
+  "data": {
+    "bindTo": "my-source",
+    "ok": false,
+    "result": null,
+    "issues": ["Probe failed: network timeout"]
+  }
+}
+```
+
+Field notes:
+
+- `result` is parsed JSON when possible; otherwise it is returned as a raw string
+- `issues` is empty on success and contains live-fetch errors when `ok` is `false`
+- If no task executor is configured, the call fails with an HTTP error such as `{ "error": "No task-executor registered for this board" }`
+- Request-shape problems such as an out-of-range `source_idx` also fail with an HTTP error, for example `{ "error": "sourceIdx 4 out of range (card has 1 source(s))" }`
+
+---
+
+### `preflight.run-single-source-in-live-card`
+
+Executes a single source using an already-saved live card in card storage. This is useful when you want run-source behavior without supplying `candidate_card_content`.
+
+**Args:**
+
+| Field | Type | Notes |
+|---|---|---|
+| `card_id` | string | required; id of an existing live card |
+| `source_idx` | number | required; index into the card's `source_defs` array |
+| `mock_requires` | object | required; same contract as `preflight.run-one-cycle-with-candidate-card` |
+
+**MCP request body:**
+```json
+{
+  "tool": "preflight.run-single-source-in-live-card",
+  "args": {
+    "card_id": "card-market-prices",
+    "source_idx": 0,
+    "mock_requires": {}
+  }
+}
+```
+
+**Returns:** same shape as `preflight.run-single-source-in-candidate-card`:
+```json
+{
+  "status": "success",
+  "data": {
+    "bindTo": "my-source",
+    "ok": true,
+    "result": { "items": [ ... ] },
+    "issues": []
+  }
+}
+```
+
+On execution failure this still returns `status: "success"` with `ok: false` and `issues` populated.
+
+```json
+{
+  "status": "success",
+  "data": {
+    "bindTo": "my-source",
+    "ok": false,
+    "result": null,
+    "issues": ["Probe failed: network timeout"]
+  }
+}
+```
+
+Request/config errors fail with HTTP error envelopes, for example:
+
+- `{ "error": "Card \"missing-card\" not found" }`
+- `{ "error": "sourceIdx 4 out of range (card has 1 source(s))" }`
+- `{ "error": "MCP tool requires mock_requires" }`
+
+---
+
+### `preflight.run-one-cycle-with-candidate-card`
+
+Runs a full computation cycle (all sources → compute) against a candidate card in-memory and returns the simplified MCP-facing runtime shape. Nothing is persisted.
+
+**Args:**
+
+| Field | Type | Notes |
+|---|---|---|
+| `candidate_card_content` | object | required |
+| `mock_requires` | object | optional mock values for `requires` keys |
+
+**MCP request body:**
+```json
+{
+  "tool": "preflight.run-one-cycle-with-candidate-card",
+  "args": {
+    "candidate_card_content": { "id": "my-card", "card_data": {}, "provides": [], "view": {} },
+    "mock_requires": {}
+  }
+}
+```
+
+**Returns:**
+```json
+{
+  "status": "success",
+  "data": {
+    "cardId": "my-card",
+    "ok": true,
+    "issues": [],
+    "provides_outputs": {
+      "output-key": { "title": "My Report", "count": 42 }
+    },
+    "rendered_view": {
+      "layout": "stack",
+      "features": {},
+      "elements": [
+        { "id": "summary", "kind": "text", "label": "Summary", "visible": true, "resolved": "My Report" }
+      ]
+    }
+  }
+}
+```
+
+Field notes:
+
+- `issues` is a flattened list built from validation issues, source probe errors, projection errors, and compute errors
+- `provides_outputs` is resolved from the card's public `provides[]` bindings; if no `provides[]` is declared, the default binding is `{ "bindTo": cardId, "ref": "card_data" }`
+- `rendered_view` is materialized from the simulated runtime node using the card's `view`
+- The MCP facade intentionally hides low-level `validation`, `source_probes`, `projection_errors`, and `compute_errors` details behind `ok` + `issues`
+
+---
+
+

package/examples/board/server/board-server.js

@@ -16,6 +16,7 @@ import {
 import {
   buildLocalBaseSpec,
   createFsBoardPlatformAdapter,
+  createFsBoardNonCorePlatformAdapter,
   createFsBoardChatStorage,
   createNodeSpawnInvocationAdapter,
   createArtifactsStore,
@@ -301,7 +302,9 @@ function buildBoardContextConfig(label, boardDir, taskExecPath, chatHandlerFlow,
   const notifyChannel = `yaml-flow-server-${label}-${boardId}-${process.pid}`;
   const baseRef = parseRef(serializeRef({ kind: 'fs-path', value: boardDir }));
   const boardAdapter = createFsBoardPlatformAdapter(baseRef, { notifyChannel });
+  const nonCoreAdapter = createFsBoardNonCorePlatformAdapter(baseRef, { notifyChannel });
   boardAdapter.requestProcessAccumulated = () => {};
+  nonCoreAdapter.requestProcessAccumulated = () => {};
 
   const artifactsRef = parseRef(serializeRef({ kind: 'fs-path', value: runtimeCardsDir }));
   const artifactsAdapter = createFsBoardPlatformAdapter(artifactsRef, { suppressSpawn: true });
@@ -313,6 +316,7 @@ function buildBoardContextConfig(label, boardDir, taskExecPath, chatHandlerFlow,
   return {
     label,
     boardAdapter,
+    nonCoreAdapter,
     artifactsAdapter,
     baseRef,
     cardStoreRef,