chore: bump version to 1.4.1 and update changelog

Document the reference column filter support feature.

.github/workflows/build.yaml

@@ -18,7 +18,7 @@ jobs:
 
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6
 
       - name: Log in to Container Registry
         uses: docker/login-action@v3

CHANGELOG.md

@@ -5,6 +5,107 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.4.1] - 2026-01-14
+
+### Added
+
+#### Reference Column Filter Support
+- **Filter normalization**: `get_records` now automatically normalizes filter values to array format
+- Fixes 400 errors when filtering on `Ref:*` (reference/foreign key) columns
+- Single values are wrapped in arrays before sending to Grist API
+
+#### Usage
+```python
+# Before: Failed with 400 Bad Request
+get_records(document="accounting", table="TransactionLines", filter={"Transaction": 44})
+
+# After: Works - filter normalized to {"Transaction": [44]}
+get_records(document="accounting", table="TransactionLines", filter={"Transaction": 44})
+
+# Multiple values also supported
+get_records(document="accounting", table="TransactionLines", filter={"Transaction": [44, 45, 46]})
+```
+
+### Fixed
+- Shell script shebangs updated to `#!/usr/bin/env bash` for portability across environments
+
+## [1.4.0] - 2026-01-12
+
+### Added
+
+#### Attachment Download via Proxy
+- **`GET /api/v1/attachments/{id}`**: New HTTP endpoint for downloading attachments
+- Returns binary content with appropriate `Content-Type` and `Content-Disposition` headers
+- Requires read permission in session token
+- Complements the existing upload endpoint for complete attachment workflows
+
+#### Usage
+```bash
+# Get session token with read permission
+TOKEN=$(curl -s ... | jq -r '.token')
+
+# Download attachment
+curl -H "Authorization: Bearer $TOKEN" \
+  https://example.com/api/v1/attachments/42 \
+  -o downloaded.pdf
+```
+
+```python
+# Python example
+import requests
+
+response = requests.get(
+    f'{base_url}/api/v1/attachments/42',
+    headers={'Authorization': f'Bearer {token}'}
+)
+with open('downloaded.pdf', 'wb') as f:
+    f.write(response.content)
+```
+
+## [1.3.0] - 2026-01-03
+
+### Added
+
+#### Attachment Upload via Proxy
+- **`POST /api/v1/attachments`**: New HTTP endpoint for file uploads
+- Uses `multipart/form-data` for efficient binary transfer (no base64 overhead)
+- Automatic MIME type detection from filename
+- Returns attachment ID for linking to records via `update_records`
+- Requires write permission in session token
+
+#### Usage
+```bash
+# Get session token with write permission
+TOKEN=$(curl -s ... | jq -r '.token')
+
+# Upload file
+curl -X POST \
+  -H "Authorization: Bearer $TOKEN" \
+  -F "file=@invoice.pdf" \
+  https://example.com/api/v1/attachments
+
+# Returns: {"success": true, "data": {"attachment_id": 42, "filename": "invoice.pdf", "size_bytes": 31395}}
+```
+
+```python
+# Python example
+import requests
+
+response = requests.post(
+    f'{proxy_url.replace("/proxy", "/attachments")}',
+    headers={'Authorization': f'Bearer {token}'},
+    files={'file': open('invoice.pdf', 'rb')}
+)
+attachment_id = response.json()['data']['attachment_id']
+
+# Link to record via proxy
+requests.post(proxy_url, headers={'Authorization': f'Bearer {token}'}, json={
+    'method': 'update_records',
+    'table': 'Bills',
+    'records': [{'id': 1, 'fields': {'Attachment': [attachment_id]}}]
+})
+```
+
 ## [1.2.0] - 2026-01-02
 
 ### Added

docs/plans/2026-01-03-attachment-upload-design.md

@@ -0,0 +1,187 @@
+# Attachment Upload Feature Design
+
+**Date:** 2026-01-03
+**Status:** Approved
+
+## Summary
+
+Add an `upload_attachment` MCP tool to upload files to Grist documents and receive an attachment ID for linking to records.
+
+## Design Decisions
+
+| Decision | Choice | Rationale |
+|----------|--------|-----------|
+| Content encoding | Base64 string | MCP tools use JSON; binary must be encoded |
+| Batch support | Single file only | YAGNI; caller can loop if needed |
+| Linking behavior | Upload only, return ID | Single responsibility; use existing `update_records` to link |
+| Download support | Not included | YAGNI; can add later if needed |
+| Permission level | Write | Attachments are data, not schema |
+| Proxy support | MCP tool only | Reduces scope; scripts can use Grist API directly |
+
+## Tool Interface
+
+### Input Schema
+
+```json
+{
+  "type": "object",
+  "properties": {
+    "document": {
+      "type": "string",
+      "description": "Document name"
+    },
+    "filename": {
+      "type": "string",
+      "description": "Filename with extension (e.g., 'invoice.pdf')"
+    },
+    "content_base64": {
+      "type": "string",
+      "description": "File content as base64-encoded string"
+    },
+    "content_type": {
+      "type": "string",
+      "description": "MIME type (optional, auto-detected from filename if omitted)"
+    }
+  },
+  "required": ["document", "filename", "content_base64"]
+}
+```
+
+### Response
+
+```json
+{
+  "attachment_id": 42,
+  "filename": "invoice.pdf",
+  "size_bytes": 30720
+}
+```
+
+### Usage Example
+
+```python
+# 1. Upload attachment
+result = upload_attachment(
+    document="accounting",
+    filename="Invoice-001.pdf",
+    content_base64="JVBERi0xLjQK..."
+)
+
+# 2. Link to record via existing update_records tool
+update_records("Bills", [{
+    "id": 1,
+    "fields": {"Attachment": [result["attachment_id"]]}
+}])
+```
+
+## Implementation
+
+### Files to Modify
+
+1. **`src/grist_mcp/grist_client.py`** - Add `upload_attachment()` method
+2. **`src/grist_mcp/tools/write.py`** - Add tool function
+3. **`src/grist_mcp/server.py`** - Register tool
+
+### GristClient Method
+
+```python
+async def upload_attachment(
+    self,
+    filename: str,
+    content: bytes,
+    content_type: str | None = None
+) -> dict:
+    """Upload a file attachment. Returns attachment metadata."""
+    if content_type is None:
+        content_type = "application/octet-stream"
+
+    files = {"upload": (filename, content, content_type)}
+
+    async with httpx.AsyncClient(timeout=self._timeout) as client:
+        response = await client.post(
+            f"{self._base_url}/attachments",
+            headers=self._headers,
+            files=files,
+        )
+        response.raise_for_status()
+        # Grist returns list of attachment IDs
+        attachment_ids = response.json()
+        return {
+            "attachment_id": attachment_ids[0],
+            "filename": filename,
+            "size_bytes": len(content),
+        }
+```
+
+### Tool Function
+
+```python
+import base64
+import mimetypes
+
+async def upload_attachment(
+    agent: Agent,
+    auth: Authenticator,
+    document: str,
+    filename: str,
+    content_base64: str,
+    content_type: str | None = None,
+    client: GristClient | None = None,
+) -> dict:
+    """Upload a file attachment to a document."""
+    auth.authorize(agent, document, Permission.WRITE)
+
+    # Decode base64
+    try:
+        content = base64.b64decode(content_base64)
+    except Exception:
+        raise ValueError("Invalid base64 encoding")
+
+    # Auto-detect MIME type if not provided
+    if content_type is None:
+        content_type, _ = mimetypes.guess_type(filename)
+        if content_type is None:
+            content_type = "application/octet-stream"
+
+    if client is None:
+        doc = auth.get_document(document)
+        client = GristClient(doc)
+
+    return await client.upload_attachment(filename, content, content_type)
+```
+
+## Error Handling
+
+| Error | Cause | Response |
+|-------|-------|----------|
+| Invalid base64 | Malformed content_base64 | `ValueError: Invalid base64 encoding` |
+| Authorization | Agent lacks write permission | `AuthError` (existing pattern) |
+| Grist API error | Upload fails | `httpx.HTTPStatusError` (existing pattern) |
+
+## Testing
+
+### Unit Tests
+
+**`tests/unit/test_tools_write.py`:**
+- `test_upload_attachment_success` - Valid base64, returns attachment_id
+- `test_upload_attachment_invalid_base64` - Raises ValueError
+- `test_upload_attachment_auth_required` - Verifies write permission check
+- `test_upload_attachment_mime_detection` - Auto-detects type from filename
+
+**`tests/unit/test_grist_client.py`:**
+- `test_upload_attachment_api_call` - Correct multipart request format
+- `test_upload_attachment_with_explicit_content_type` - Passes through MIME type
+
+### Mock Approach
+
+Mock `httpx.AsyncClient` responses; no Grist server needed for unit tests.
+
+## Future Considerations
+
+Not included in this implementation (YAGNI):
+- Batch upload (multiple files)
+- Download attachment
+- Proxy API support
+- Size limit validation (rely on Grist's limits)
+
+These can be added if real use cases emerge.

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "grist-mcp"
-version = "1.2.0"
+version = "1.4.1"
 description = "MCP server for AI agents to interact with Grist documents"
 requires-python = ">=3.14"
 dependencies = [

scripts/get-test-instance-id.sh

@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 # scripts/get-test-instance-id.sh
 # Generate a unique instance ID from git branch for parallel test isolation
 

scripts/run-integration-tests.sh

@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 # scripts/run-integration-tests.sh
 # Run integration tests with branch isolation and dynamic port discovery
 set -e

src/grist_mcp/grist_client.py

@@ -116,6 +116,71 @@ class GristClient:
         """Delete records by ID."""
         await self._request("POST", f"/tables/{table}/data/delete", json=record_ids)
 
+    async def upload_attachment(
+        self,
+        filename: str,
+        content: bytes,
+        content_type: str = "application/octet-stream",
+    ) -> dict:
+        """Upload a file attachment. Returns attachment metadata.
+
+        Args:
+            filename: Name for the uploaded file.
+            content: File content as bytes.
+            content_type: MIME type of the file.
+
+        Returns:
+            Dict with attachment_id, filename, and size_bytes.
+        """
+        files = {"upload": (filename, content, content_type)}
+
+        async with httpx.AsyncClient(timeout=self._timeout) as client:
+            response = await client.post(
+                f"{self._base_url}/attachments",
+                headers=self._headers,
+                files=files,
+            )
+            response.raise_for_status()
+            # Grist returns list of attachment IDs
+            attachment_ids = response.json()
+            return {
+                "attachment_id": attachment_ids[0],
+                "filename": filename,
+                "size_bytes": len(content),
+            }
+
+    async def download_attachment(self, attachment_id: int) -> dict:
+        """Download an attachment by ID.
+
+        Args:
+            attachment_id: The ID of the attachment to download.
+
+        Returns:
+            Dict with content (bytes), content_type, and filename.
+        """
+        import re
+
+        async with httpx.AsyncClient(timeout=self._timeout) as client:
+            response = await client.get(
+                f"{self._base_url}/attachments/{attachment_id}/download",
+                headers=self._headers,
+            )
+            response.raise_for_status()
+
+            # Extract filename from Content-Disposition header
+            content_disp = response.headers.get("content-disposition", "")
+            filename = None
+            if "filename=" in content_disp:
+                match = re.search(r'filename="?([^";]+)"?', content_disp)
+                if match:
+                    filename = match.group(1)
+
+            return {
+                "content": response.content,
+                "content_type": response.headers.get("content-type", "application/octet-stream"),
+                "filename": filename,
+            }
+
     # Schema operations
 
     async def create_table(self, table_id: str, columns: list[dict]) -> str:

src/grist_mcp/main.py

@@ -14,6 +14,7 @@ from grist_mcp.config import Config, load_config
 from grist_mcp.auth import Authenticator, AuthError
 from grist_mcp.session import SessionTokenManager
 from grist_mcp.proxy import parse_proxy_request, dispatch_proxy_request, ProxyError
+from grist_mcp.grist_client import GristClient
 from grist_mcp.logging import setup_logging
 
 
@@ -59,6 +60,62 @@ async def send_json_response(send: Send, status: int, data: dict) -> None:
     })
 
 
+def _parse_multipart(content_type: str, body: bytes) -> tuple[str | None, bytes | None]:
+    """Parse multipart/form-data to extract uploaded file.
+
+    Returns (filename, content) or (None, None) if parsing fails.
+    """
+    import re
+
+    # Extract boundary from content-type
+    match = re.search(r'boundary=([^\s;]+)', content_type)
+    if not match:
+        return None, None
+
+    boundary = match.group(1).encode()
+    if boundary.startswith(b'"') and boundary.endswith(b'"'):
+        boundary = boundary[1:-1]
+
+    # Split by boundary
+    parts = body.split(b'--' + boundary)
+
+    for part in parts:
+        if b'Content-Disposition' not in part:
+            continue
+
+        # Split headers from content
+        if b'\r\n\r\n' in part:
+            header_section, content = part.split(b'\r\n\r\n', 1)
+        elif b'\n\n' in part:
+            header_section, content = part.split(b'\n\n', 1)
+        else:
+            continue
+
+        headers = header_section.decode('utf-8', errors='replace')
+
+        # Check if this is a file upload
+        if 'filename=' not in headers:
+            continue
+
+        # Extract filename
+        filename_match = re.search(r'filename="([^"]+)"', headers)
+        if not filename_match:
+            filename_match = re.search(r"filename=([^\s;]+)", headers)
+        if not filename_match:
+            continue
+
+        filename = filename_match.group(1)
+
+        # Remove trailing boundary marker and whitespace
+        content = content.rstrip()
+        if content.endswith(b'--'):
+            content = content[:-2].rstrip()
+
+        return filename, content
+
+    return None, None
+
+
 CONFIG_TEMPLATE = """\
 # grist-mcp configuration
 #
@@ -229,6 +286,144 @@ def create_app(config: Config):
                 "code": e.code,
             })
 
+    async def handle_attachments(scope: Scope, receive: Receive, send: Send) -> None:
+        """Handle file attachment uploads via multipart/form-data."""
+        # Extract token
+        token = _get_bearer_token(scope)
+        if not token:
+            await send_json_response(send, 401, {
+                "success": False,
+                "error": "Missing Authorization header",
+                "code": "INVALID_TOKEN",
+            })
+            return
+
+        # Validate session token
+        session = token_manager.validate_token(token)
+        if session is None:
+            await send_json_response(send, 401, {
+                "success": False,
+                "error": "Invalid or expired token",
+                "code": "TOKEN_EXPIRED",
+            })
+            return
+
+        # Check write permission
+        if "write" not in session.permissions:
+            await send_json_response(send, 403, {
+                "success": False,
+                "error": "Write permission required for attachment upload",
+                "code": "UNAUTHORIZED",
+            })
+            return
+
+        # Get content-type header
+        headers = dict(scope.get("headers", []))
+        content_type = headers.get(b"content-type", b"").decode()
+
+        if not content_type.startswith("multipart/form-data"):
+            await send_json_response(send, 400, {
+                "success": False,
+                "error": "Content-Type must be multipart/form-data",
+                "code": "INVALID_REQUEST",
+            })
+            return
+
+        # Read request body
+        body = b""
+        while True:
+            message = await receive()
+            body += message.get("body", b"")
+            if not message.get("more_body", False):
+                break
+
+        # Parse multipart
+        filename, content = _parse_multipart(content_type, body)
+        if filename is None or content is None:
+            await send_json_response(send, 400, {
+                "success": False,
+                "error": "No file found in request",
+                "code": "INVALID_REQUEST",
+            })
+            return
+
+        # Upload to Grist
+        try:
+            doc = auth.get_document(session.document)
+            client = GristClient(doc)
+            result = await client.upload_attachment(filename, content)
+            await send_json_response(send, 200, {
+                "success": True,
+                "data": result,
+            })
+        except Exception as e:
+            await send_json_response(send, 500, {
+                "success": False,
+                "error": str(e),
+                "code": "GRIST_ERROR",
+            })
+
+    async def handle_attachment_download(
+        scope: Scope, receive: Receive, send: Send, attachment_id: int
+    ) -> None:
+        """Handle attachment download by ID."""
+        # Extract token
+        token = _get_bearer_token(scope)
+        if not token:
+            await send_json_response(send, 401, {
+                "success": False,
+                "error": "Missing Authorization header",
+                "code": "INVALID_TOKEN",
+            })
+            return
+
+        # Validate session token
+        session = token_manager.validate_token(token)
+        if session is None:
+            await send_json_response(send, 401, {
+                "success": False,
+                "error": "Invalid or expired token",
+                "code": "TOKEN_EXPIRED",
+            })
+            return
+
+        # Check read permission
+        if "read" not in session.permissions:
+            await send_json_response(send, 403, {
+                "success": False,
+                "error": "Read permission required for attachment download",
+                "code": "UNAUTHORIZED",
+            })
+            return
+
+        # Download from Grist
+        try:
+            doc = auth.get_document(session.document)
+            client = GristClient(doc)
+            result = await client.download_attachment(attachment_id)
+
+            # Build response headers
+            headers = [[b"content-type", result["content_type"].encode()]]
+            if result["filename"]:
+                disposition = f'attachment; filename="{result["filename"]}"'
+                headers.append([b"content-disposition", disposition.encode()])
+
+            await send({
+                "type": "http.response.start",
+                "status": 200,
+                "headers": headers,
+            })
+            await send({
+                "type": "http.response.body",
+                "body": result["content"],
+            })
+        except Exception as e:
+            await send_json_response(send, 500, {
+                "success": False,
+                "error": str(e),
+                "code": "GRIST_ERROR",
+            })
+
     async def app(scope: Scope, receive: Receive, send: Send) -> None:
         if scope["type"] != "http":
             return
@@ -244,6 +439,19 @@ def create_app(config: Config):
             await handle_messages(scope, receive, send)
         elif path == "/api/v1/proxy" and method == "POST":
             await handle_proxy(scope, receive, send)
+        elif path == "/api/v1/attachments" and method == "POST":
+            await handle_attachments(scope, receive, send)
+        elif path.startswith("/api/v1/attachments/") and method == "GET":
+            # Parse attachment ID from path: /api/v1/attachments/{id}
+            try:
+                attachment_id = int(path.split("/")[-1])
+                await handle_attachment_download(scope, receive, send, attachment_id)
+            except ValueError:
+                await send_json_response(send, 400, {
+                    "success": False,
+                    "error": "Invalid attachment ID",
+                    "code": "INVALID_REQUEST",
+                })
         else:
             await handle_not_found(scope, receive, send)
 

src/grist_mcp/tools/filters.py

@@ -0,0 +1,37 @@
+"""Filter normalization for Grist API queries."""
+
+from typing import Any
+
+
+def normalize_filter_value(value: Any) -> list:
+    """Ensure a filter value is a list.
+
+    Grist API expects filter values to be arrays.
+
+    Args:
+        value: Single value or list of values.
+
+    Returns:
+        Value wrapped in list, or original list if already a list.
+    """
+    if isinstance(value, list):
+        return value
+    return [value]
+
+
+def normalize_filter(filter: dict | None) -> dict | None:
+    """Normalize filter values to array format for Grist API.
+
+    Grist expects all filter values to be arrays. This function
+    wraps single values in lists.
+
+    Args:
+        filter: Filter dict with column names as keys.
+
+    Returns:
+        Normalized filter dict, or None if input was None.
+    """
+    if not filter:
+        return filter
+
+    return {key: normalize_filter_value(value) for key, value in filter.items()}

src/grist_mcp/tools/read.py

@@ -2,6 +2,7 @@
 
 from grist_mcp.auth import Agent, Authenticator, Permission
 from grist_mcp.grist_client import GristClient
+from grist_mcp.tools.filters import normalize_filter
 
 
 async def list_tables(
@@ -56,7 +57,10 @@ async def get_records(
         doc = auth.get_document(document)
         client = GristClient(doc)
 
-    records = await client.get_records(table, filter=filter, sort=sort, limit=limit)
+    # Normalize filter values to array format for Grist API
+    normalized_filter = normalize_filter(filter)
+
+    records = await client.get_records(table, filter=normalized_filter, sort=sort, limit=limit)
     return {"records": records}
 
 

src/grist_mcp/tools/session.py

@@ -6,9 +6,43 @@ from grist_mcp.session import SessionTokenManager
 
 PROXY_DOCUMENTATION = {
     "description": "HTTP proxy API for bulk data operations. Use request_session_token to get a short-lived token, then call the proxy endpoint directly from scripts.",
-    "endpoint": "POST /api/v1/proxy",
-    "endpoint_note": "The full URL is returned in the 'proxy_url' field of request_session_token response",
+    "endpoints": {
+        "proxy": "POST /api/v1/proxy - JSON operations (CRUD, schema)",
+        "attachments_upload": "POST /api/v1/attachments - File uploads (multipart/form-data)",
+        "attachments_download": "GET /api/v1/attachments/{id} - File downloads (binary response)",
+    },
+    "endpoint_note": "The full URL is returned in the 'proxy_url' field of request_session_token response. Replace /proxy with /attachments for file operations.",
     "authentication": "Bearer token in Authorization header",
+    "attachment_upload": {
+        "endpoint": "POST /api/v1/attachments",
+        "content_type": "multipart/form-data",
+        "permission": "write",
+        "description": "Upload file attachments to the document. Returns attachment_id for linking to records via update_records.",
+        "response": {"success": True, "data": {"attachment_id": 42, "filename": "invoice.pdf", "size_bytes": 31395}},
+        "example_curl": "curl -X POST -H 'Authorization: Bearer TOKEN' -F 'file=@invoice.pdf' URL/api/v1/attachments",
+        "example_python": """import requests
+response = requests.post(
+    f'{proxy_url.replace("/proxy", "/attachments")}',
+    headers={'Authorization': f'Bearer {token}'},
+    files={'file': open('invoice.pdf', 'rb')}
+)
+attachment_id = response.json()['data']['attachment_id']
+# Link to record: update_records with {'Attachment': [attachment_id]}""",
+    },
+    "attachment_download": {
+        "endpoint": "GET /api/v1/attachments/{attachment_id}",
+        "permission": "read",
+        "description": "Download attachment by ID. Returns binary content with appropriate Content-Type and Content-Disposition headers.",
+        "response_headers": ["Content-Type", "Content-Disposition"],
+        "example_curl": "curl -H 'Authorization: Bearer TOKEN' URL/api/v1/attachments/42 -o file.pdf",
+        "example_python": """import requests
+response = requests.get(
+    f'{base_url}/api/v1/attachments/42',
+    headers={'Authorization': f'Bearer {token}'}
+)
+with open('downloaded.pdf', 'wb') as f:
+    f.write(response.content)""",
+    },
     "request_format": {
         "method": "Operation name (required)",
         "table": "Table name (required for most operations)",

tests/integration/mock_grist/server.py

@@ -35,6 +35,18 @@ MOCK_TABLES = {
             {"id": 2, "fields": {"Title": "Deploy", "Done": False}},
         ],
     },
+    "Orders": {
+        "columns": [
+            {"id": "OrderNum", "fields": {"type": "Int"}},
+            {"id": "Customer", "fields": {"type": "Ref:People"}},
+            {"id": "Amount", "fields": {"type": "Numeric"}},
+        ],
+        "records": [
+            {"id": 1, "fields": {"OrderNum": 1001, "Customer": 1, "Amount": 100.0}},
+            {"id": 2, "fields": {"OrderNum": 1002, "Customer": 2, "Amount": 200.0}},
+            {"id": 3, "fields": {"OrderNum": 1003, "Customer": 1, "Amount": 150.0}},
+        ],
+    },
 }
 
 # Track requests for test assertions
@@ -93,12 +105,40 @@ async def get_records(request):
     """GET /api/docs/{doc_id}/tables/{table_id}/records"""
     doc_id = request.path_params["doc_id"]
     table_id = request.path_params["table_id"]
-    log_request("GET", f"/api/docs/{doc_id}/tables/{table_id}/records")
+    filter_param = request.query_params.get("filter")
+    log_request("GET", f"/api/docs/{doc_id}/tables/{table_id}/records?filter={filter_param}")
 
     if table_id not in MOCK_TABLES:
         return JSONResponse({"error": "Table not found"}, status_code=404)
 
-    return JSONResponse({"records": MOCK_TABLES[table_id]["records"]})
+    records = MOCK_TABLES[table_id]["records"]
+
+    # Apply filtering if provided
+    if filter_param:
+        try:
+            filters = json.loads(filter_param)
+            # Validate filter format: all values must be arrays (Grist API requirement)
+            for key, values in filters.items():
+                if not isinstance(values, list):
+                    return JSONResponse(
+                        {"error": f"Filter values must be arrays, got {type(values).__name__} for '{key}'"},
+                        status_code=400
+                    )
+            # Apply filters: record matches if field value is in the filter list
+            filtered_records = []
+            for record in records:
+                match = True
+                for key, allowed_values in filters.items():
+                    if record["fields"].get(key) not in allowed_values:
+                        match = False
+                        break
+                if match:
+                    filtered_records.append(record)
+            records = filtered_records
+        except json.JSONDecodeError:
+            return JSONResponse({"error": "Invalid filter JSON"}, status_code=400)
+
+    return JSONResponse({"records": records})
 
 
 async def add_records(request):

tests/integration/test_tools_integration.py

@@ -90,6 +90,36 @@ async def test_all_tools(services_ready):
         log = get_mock_request_log()
         assert any("/records" in entry["path"] and entry["method"] == "GET" for entry in log)
 
+        # Test get_records with Ref column filter
+        # This tests that single values are normalized to arrays for the Grist API
+        clear_mock_request_log()
+        result = await client.call_tool(
+            "get_records",
+            {"document": "test-doc", "table": "Orders", "filter": {"Customer": 1}}
+        )
+        data = json.loads(result.content[0].text)
+        assert "records" in data
+        # Should return only orders for Customer 1 (orders 1 and 3)
+        assert len(data["records"]) == 2
+        for record in data["records"]:
+            assert record["Customer"] == 1
+        log = get_mock_request_log()
+        # Verify the filter was sent as array format
+        filter_requests = [e for e in log if "/records" in e["path"] and "filter=" in e["path"]]
+        assert len(filter_requests) >= 1
+        # The filter value should be [1] not 1
+        assert "[1]" in filter_requests[0]["path"]
+
+        # Test get_records with multiple filter values
+        clear_mock_request_log()
+        result = await client.call_tool(
+            "get_records",
+            {"document": "test-doc", "table": "Orders", "filter": {"Customer": [1, 2]}}
+        )
+        data = json.loads(result.content[0].text)
+        assert "records" in data
+        assert len(data["records"]) == 3  # All 3 orders (customers 1 and 2)
+
         # Test sql_query
         clear_mock_request_log()
         result = await client.call_tool(

tests/unit/test_filters.py

@@ -0,0 +1,89 @@
+"""Unit tests for filter normalization."""
+
+import pytest
+
+from grist_mcp.tools.filters import normalize_filter, normalize_filter_value
+
+
+class TestNormalizeFilterValue:
+    """Tests for normalize_filter_value function."""
+
+    def test_int_becomes_list(self):
+        assert normalize_filter_value(5) == [5]
+
+    def test_string_becomes_list(self):
+        assert normalize_filter_value("foo") == ["foo"]
+
+    def test_float_becomes_list(self):
+        assert normalize_filter_value(3.14) == [3.14]
+
+    def test_list_unchanged(self):
+        assert normalize_filter_value([1, 2, 3]) == [1, 2, 3]
+
+    def test_empty_list_unchanged(self):
+        assert normalize_filter_value([]) == []
+
+    def test_single_item_list_unchanged(self):
+        assert normalize_filter_value([42]) == [42]
+
+    def test_mixed_type_list_unchanged(self):
+        assert normalize_filter_value([1, "foo", 3.14]) == [1, "foo", 3.14]
+
+
+class TestNormalizeFilter:
+    """Tests for normalize_filter function."""
+
+    def test_none_returns_none(self):
+        assert normalize_filter(None) is None
+
+    def test_empty_dict_returns_empty_dict(self):
+        assert normalize_filter({}) == {}
+
+    def test_single_int_value_wrapped(self):
+        result = normalize_filter({"Transaction": 44})
+        assert result == {"Transaction": [44]}
+
+    def test_single_string_value_wrapped(self):
+        result = normalize_filter({"Status": "active"})
+        assert result == {"Status": ["active"]}
+
+    def test_list_value_unchanged(self):
+        result = normalize_filter({"Transaction": [44, 45, 46]})
+        assert result == {"Transaction": [44, 45, 46]}
+
+    def test_mixed_columns_all_normalized(self):
+        """Both ref and non-ref columns are normalized to arrays."""
+        result = normalize_filter({
+            "Transaction": 44,  # Ref column (int)
+            "Debit": 500,       # Non-ref column (int)
+            "Memo": "test",     # Non-ref column (str)
+        })
+        assert result == {
+            "Transaction": [44],
+            "Debit": [500],
+            "Memo": ["test"],
+        }
+
+    def test_multiple_values_list_unchanged(self):
+        """Filter with multiple values passes through."""
+        result = normalize_filter({
+            "Status": ["pending", "active"],
+            "Priority": [1, 2, 3],
+        })
+        assert result == {
+            "Status": ["pending", "active"],
+            "Priority": [1, 2, 3],
+        }
+
+    def test_mixed_single_and_list_values(self):
+        """Mix of single values and lists."""
+        result = normalize_filter({
+            "Transaction": 44,            # Single int
+            "Status": ["open", "closed"],  # List
+            "Amount": 100.50,             # Single float
+        })
+        assert result == {
+            "Transaction": [44],
+            "Status": ["open", "closed"],
+            "Amount": [100.50],
+        }

tests/unit/test_grist_client.py

@@ -196,3 +196,99 @@ def test_sql_validation_rejects_multiple_statements(client):
 def test_sql_validation_allows_trailing_semicolon(client):
     # Should not raise
     client._validate_sql_query("SELECT * FROM users;")
+
+
+# Attachment tests
+
+@pytest.mark.asyncio
+async def test_upload_attachment(client, httpx_mock: HTTPXMock):
+    httpx_mock.add_response(
+        url="https://grist.example.com/api/docs/abc123/attachments",
+        method="POST",
+        json=[42],
+    )
+
+    result = await client.upload_attachment(
+        filename="invoice.pdf",
+        content=b"PDF content here",
+        content_type="application/pdf",
+    )
+
+    assert result == {
+        "attachment_id": 42,
+        "filename": "invoice.pdf",
+        "size_bytes": 16,
+    }
+
+
+@pytest.mark.asyncio
+async def test_upload_attachment_default_content_type(client, httpx_mock: HTTPXMock):
+    httpx_mock.add_response(
+        url="https://grist.example.com/api/docs/abc123/attachments",
+        method="POST",
+        json=[99],
+    )
+
+    result = await client.upload_attachment(
+        filename="data.bin",
+        content=b"\x00\x01\x02",
+    )
+
+    assert result["attachment_id"] == 99
+    assert result["size_bytes"] == 3
+
+
+@pytest.mark.asyncio
+async def test_download_attachment(client, httpx_mock: HTTPXMock):
+    httpx_mock.add_response(
+        url="https://grist.example.com/api/docs/abc123/attachments/42/download",
+        method="GET",
+        content=b"PDF content here",
+        headers={
+            "content-type": "application/pdf",
+            "content-disposition": 'attachment; filename="invoice.pdf"',
+        },
+    )
+
+    result = await client.download_attachment(42)
+
+    assert result["content"] == b"PDF content here"
+    assert result["content_type"] == "application/pdf"
+    assert result["filename"] == "invoice.pdf"
+
+
+@pytest.mark.asyncio
+async def test_download_attachment_no_filename(client, httpx_mock: HTTPXMock):
+    httpx_mock.add_response(
+        url="https://grist.example.com/api/docs/abc123/attachments/99/download",
+        method="GET",
+        content=b"binary data",
+        headers={
+            "content-type": "application/octet-stream",
+        },
+    )
+
+    result = await client.download_attachment(99)
+
+    assert result["content"] == b"binary data"
+    assert result["content_type"] == "application/octet-stream"
+    assert result["filename"] is None
+
+
+@pytest.mark.asyncio
+async def test_download_attachment_unquoted_filename(client, httpx_mock: HTTPXMock):
+    httpx_mock.add_response(
+        url="https://grist.example.com/api/docs/abc123/attachments/55/download",
+        method="GET",
+        content=b"image data",
+        headers={
+            "content-type": "image/png",
+            "content-disposition": "attachment; filename=photo.png",
+        },
+    )
+
+    result = await client.download_attachment(55)
+
+    assert result["content"] == b"image data"
+    assert result["content_type"] == "image/png"
+    assert result["filename"] == "photo.png"

tests/unit/test_tools_read.py

@@ -75,6 +75,45 @@ async def test_get_records(agent, auth, mock_client):
     assert result == {"records": [{"id": 1, "Name": "Alice"}]}
 
 
+@pytest.mark.asyncio
+async def test_get_records_normalizes_filter(agent, auth, mock_client):
+    """Test that filter values are normalized to array format for Grist API."""
+    mock_client.get_records.return_value = [{"id": 1, "Customer": 5}]
+
+    await get_records(
+        agent, auth, "budget", "Orders",
+        filter={"Customer": 5, "Status": "active"},
+        client=mock_client,
+    )
+
+    # Verify filter was normalized: single values wrapped in lists
+    mock_client.get_records.assert_called_once_with(
+        "Orders",
+        filter={"Customer": [5], "Status": ["active"]},
+        sort=None,
+        limit=None,
+    )
+
+
+@pytest.mark.asyncio
+async def test_get_records_preserves_list_filter(agent, auth, mock_client):
+    """Test that filter values already in list format are preserved."""
+    mock_client.get_records.return_value = []
+
+    await get_records(
+        agent, auth, "budget", "Orders",
+        filter={"Customer": [5, 6, 7]},
+        client=mock_client,
+    )
+
+    mock_client.get_records.assert_called_once_with(
+        "Orders",
+        filter={"Customer": [5, 6, 7]},
+        sort=None,
+        limit=None,
+    )
+
+
 @pytest.mark.asyncio
 async def test_sql_query(agent, auth, mock_client):
     result = await sql_query(agent, auth, "budget", "SELECT * FROM Table1", client=mock_client)

tests/unit/test_tools_session.py

@@ -39,12 +39,16 @@ async def test_get_proxy_documentation_returns_complete_spec():
     result = await get_proxy_documentation()
 
     assert "description" in result
-    assert "endpoint" in result
-    assert result["endpoint"] == "POST /api/v1/proxy"
+    assert "endpoints" in result
+    assert "proxy" in result["endpoints"]
+    assert "attachments_upload" in result["endpoints"]
+    assert "attachments_download" in result["endpoints"]
     assert "authentication" in result
     assert "methods" in result
     assert "add_records" in result["methods"]
     assert "get_records" in result["methods"]
+    assert "attachment_upload" in result
+    assert "attachment_download" in result
     assert "example_script" in result
 
 

uv.lock

@@ -153,7 +153,7 @@ wheels = [
 
 [[package]]
 name = "grist-mcp"
-version = "1.2.0"
+version = "1.3.0"
 source = { editable = "." }
 dependencies = [
     { name = "httpx" },