mcp-fetch

A minimal MCP (Model Context Protocol) stdio server that exposes a single tool: http_request.

It supports:

• Any HTTP method (GET/POST/PUT/PATCH/DELETE/HEAD/OPTIONS/...)
• Headers, query params, and multiple body types (json/form/text/base64 bytes)
• Large responses via chunking with a transfer_id cursor
• In-process caching backed by on-disk files so agents can request more chunks later

Run (uvx)

From this directory:

uvx --refresh --from . mcp-fetch

This is an MCP stdio server: it typically does not print anything until an MCP client connects and sends requests. If you run it directly in a terminal, it should stay running and wait for stdin from an MCP host.

For interactive/manual testing, it will auto-run an HTTP server when stdin is a TTY:

• Open: http://127.0.0.1:8000/mcp
• Override: MCP_FETCH_TRANSPORT=stdio|http, MCP_FETCH_PORT=8000
• Or flags: mcp-fetch --transport http --port 8000

Client config note

If you use uvx, you must include the provided command name at the end. This works:

cmd: uvx.exe
args: ["--from", "E:\\Private\\Mcp\\fetch", "mcp-fetch", "--transport", "stdio"]

Tool: http_request

Typical flow:

1. Call with url (gets chunk 0 and transfer_id)
2. If you need more, call again with transfer_id and offset=next_offset

Notes:

• This server allows private/internal network access by default (your requested behavior).
• Only http:// and https:// URLs are allowed.
• Use headers for request headers.

Tests

Run unit tests:

python -m unittest discover -s tests -v

There is an integration-style test that attempts a real request to https://github.com/. If outbound network access is unavailable, it will be skipped.

Configuration (env)

• MCP_FETCH_CACHE_DIR (default: ./.mcp-fetch-cache)
• MCP_FETCH_TTL_SECONDS (default: 1800)
• MCP_FETCH_MAX_CACHE_BYTES_TOTAL (default: 536870912) (512 MiB)
• MCP_FETCH_MAX_SINGLE_TRANSFER_BYTES (default: 209715200) (200 MiB)
• MCP_FETCH_WAIT_CHUNK_TIMEOUT_SECONDS (default: 30)