Skip to content

Release v0.4.2

Choose a tag to compare

View all tags
@github-actions github-actions released this 09 Jun 13:29
v0.4.2
93be258
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Security

  • Move the Gemini TLS transport to native asyncio (asyncio.open_connection
    with ssl=), so connect, handshake and every read are genuinely cancellable:
    a slow-loris or stalled peer is now cut off at the request deadline instead of
    parking a worker thread on a blocking recv. The previous design ran blocking
    socket I/O on a thread pool shared with DNS resolution (the SSRF guard for both
    protocols), so a handful of slow Gemini reads could stall DNS for every request
    and escalate one slow server into a whole-server denial of service.
  • Stop leaking a Gemini status-11 (SENSITIVE_INPUT) answer: the percent-encoded
    query is no longer written to logs, reflected back to the model via
    requestInfo, or retained in a cache key.
  • Reject not-yet-valid certificates on TOFU first use by default (previously
    pinned with only a warning unless reject_expired was set).
  • Reject empty and self-referential Gemini redirects (INVALID_REDIRECT) so a
    malformed 3x response cannot drive an unbounded client re-fetch loop.

Added

  • GOPHER_MAX_MENU_ITEMS (default 1000): caps the number of Gopher menu items
    returned to the model, mirroring the existing text/gemtext character cap.
  • Optional positive port allowlist (GOPHER_ALLOWED_PORTS /
    GEMINI_ALLOWED_PORTS) to close the arbitrary-port port-scanning gap left by
    the dangerous-ports denylist.

Changed

  • Apply the max_rendered_chars cap to text/gemini responses (previously only
    text/*), so a large gemtext page no longer floods the model context; both
    gemtext and menu results now carry a truncated flag.
  • Drop the over-strict hardcoded TLS 1.2 cipher allow-list in favour of Python's
    secure defaults, improving interop with conforming Gemini servers.
  • De-duplicate the two batch-fetch tools (gopher_batch_fetch /
    gemini_batch_fetch) onto a single shared implementation so the batch
    error/contract behaviour has one source of truth (no behaviour change).

Removed

  • The unused http/aiohttp optional dependency, the never-read
    development_mode setting, and the test-only create_tls_client factory.
  • The GeminiGemtextResult.summary, .plain_text and .structured_content
    helper properties: they were never serialized by model_dump() so the MCP
    tools never exposed them (dead LLM-facing API). The parsed document and
    raw_content carry the same information.

Fixed

  • Correct the LLM-facing server instructions, gopher_fetch parameter
    description, and AI Assistant Guide to reference the real serialized
    next_url menu field (not the nextUrl/url names that never appear in the
    output).
  • Sync server.json to the released version and replace placeholder

Release Information:

Installation:

pip install gopher-mcp==0.4.2

Verification:
All packages are signed and can be verified using Sigstore.

Assets 4
Loading