blob-storage
A lightweight MCP data broker that stores large binary payloads under a UUID, enabling MCP clients and agents to exchange massive results without flooding the LLM context window.
README
Blob Storage Service
A lightweight MCP data broker that stores large binary payloads under a UUID, enabling MCP clients and agents to exchange massive results without flooding the LLM context window.
Architecture
MCP Client / LLM Agent
│
│ tool call (large result)
▼
MCP Server ──store_blob──▶ Blob Storage Service ──▶ SQLite + local files
│ │
│ returns UUID │ retrieve_blob(uuid)
◀──────────────────────────────┘
The agent only ever sees a UUID token — not the raw bytes.
Project Structure
src/blob_storage/
├── __init__.py # package metadata
├── config.py # pydantic-settings (env / .env)
├── models.py # SQLAlchemy ORM + Pydantic schemas
├── database.py # async engine, session factory, init_db()
├── storage.py # StorageBackend ABC + LocalFileBackend
├── service.py # BlobService (orchestration)
├── api.py # FastAPI REST API
└── mcp_server.py # MCP server (HTTP Streamable, /mcp)
tests/
├── conftest.py # shared fixtures
├── test_api.py # API integration tests
└── test_service.py # service unit tests
Quickstart
Install dependencies
uv sync
Start the REST API
uv run blob-api # defaults from config / .env
uv run blob-api --host 0.0.0.0 --port 9000 # override at runtime
Interactive docs available at http://<host>:<port>/docs.
Start the MCP server
uv run blob-mcp # defaults: 0.0.0.0:8001
uv run blob-mcp --host 0.0.0.0 --port 9001
The MCP endpoint is served at http://<host>:<port>/mcp.
Configuration
Settings are read from environment variables or a .env file in the project root.
| Variable | Default | Description |
|---|---|---|
STORAGE_DIR |
./data/blobs |
Directory for blob files |
DATABASE_URL |
sqlite+aiosqlite:///./data/metadata.db |
SQLAlchemy async DB URL |
DEFAULT_TTL_SECONDS |
7776000 |
Default TTL — 90 days. Use 0 for no expiry. |
MAX_BLOB_SIZE_BYTES |
524288000 |
Max upload size (500 MB) |
API_HOST |
0.0.0.0 |
REST API bind address |
API_PORT |
8000 |
REST API port |
MCP_HOST |
0.0.0.0 |
MCP server bind address |
MCP_PORT |
8001 |
MCP server port |
MCP_API_BASE_URL |
http://127.0.0.1:8000 |
REST API URL used internally by the MCP server |
SWEEP_INTERVAL_SECONDS |
3600 |
How often the TTL cleanup job runs |
CLI flags (
--host,--port) take precedence over config/env for the respective server.
REST API Reference
| Method | Path | Description |
|---|---|---|
GET |
/health |
Liveness check |
POST |
/blobs |
Store a blob |
GET |
/blobs/{uuid} |
Download a blob |
GET |
/blobs/{uuid}/meta |
Metadata only (no payload) |
DELETE |
/blobs/{uuid} |
Delete a blob |
Store a blob
Pass metadata as custom request headers:
| Header | Default | Description |
|---|---|---|
X-Mime-Type |
application/octet-stream |
MIME type of the payload |
X-Ttl-Seconds |
config default | Override TTL; 0 = never expires |
X-Origin |
— | Free-form source identifier |
X-Tags |
— | JSON object for arbitrary labels |
curl -X POST http://localhost:8000/blobs \
-H "Content-Type: application/octet-stream" \
-H "X-Mime-Type: application/json" \
-H "X-Origin: my-mcp-server" \
--data-binary '{"key": "value"}'
Response:
{
"uuid": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
"expires_at": "2026-05-30T16:42:07Z",
"size_bytes": 16
}
MCP Integration
The MCP server uses the Streamable HTTP transport. Configure your MCP host to connect over HTTP rather than spawning a subprocess:
{
"mcpServers": {
"blob-storage": {
"type": "http",
"url": "http://127.0.0.1:8001/mcp"
}
}
}
Available tools
| Tool | Input | Output |
|---|---|---|
store_blob |
data_b64, mime_type, ttl_seconds?, origin?, tags? |
{ uuid, expires_at, size_bytes } |
retrieve_blob |
uuid |
{ uuid, mime_type, data_b64 } |
get_blob_meta |
uuid |
metadata JSON |
delete_blob |
uuid |
{ deleted: uuid } |
All binary data is base64-encoded in MCP tool calls (MCP messages are JSON/text).
Running Tests
uv run pytest tests/ -v
Extending the Storage Backend
To use S3 or MinIO in production, implement the StorageBackend abstract class in storage.py:
from blob_storage.storage import StorageBackend
class S3Backend(StorageBackend):
async def store(self, blob_uuid: str, data: bytes) -> None: ...
async def retrieve(self, blob_uuid: str) -> bytes: ...
async def delete(self, blob_uuid: str) -> None: ...
Then inject it into BlobService and the _backend singleton in api.py.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.