Reported issues for openzim-mcp
Pod holds 10 of 10 problems reported by people outside the maintainer team. Issues filed by the project's own owners, members and collaborators are excluded entirely — a maintainer's release checklist is not a warning to a prospective user.
Back to openzim-mcp.
Most discussed
Tracking: adopt MCP spec 2026-07-28 (Python SDK v2 migration)
Context
The 2026-07-28 MCP spec revision is the "stateless MCP" release: the initialize handshake, protocol-level sessions (Mcp-Session-Id), SSE resumability (Last-Event-ID), and the HTTP GET endpoint are removed; subscriptions/listen replaces resources/subscribe/unsubscribe; a mandatory server/discover RPC replaces initialize-time capability/identity advertisement; Roots, Sampling, and Logging are deprecat
Read the thread · 2026-08-10 · closed · 6 comments
Raising the mcp ceiling past 2.0.x: seam re-audit checklist (parked — no 2.1 exists)
3.0.0 ships with mcp[cli]>=2.0.0,<2.1 — held deliberately, because the code performs surgery on 2.0.0 internals that a future 2.x could move without any public-API break. CI resolves the lockfile and stays green either way; fresh installs (pip/uvx, Smithery, Glama Docker) resolve the newest admitted wheel, which is the lockfile-masks-range-defects failure mode this project has hit before. Raising the ceiling is a deliberate bump PR that walks this checklist:
- ~~**
sdk_compat.pyping shi
Read the thread · 2026-08-20 · open · 4 comments
Re-evaluate the deferred 2026-07-28 schema capabilities (outputSchema/structuredContent, zim_search oneOf)
Split from #360 (Phase 4 optionals) so the 3.0.0 tracking issue could close. Both were deferred, not decided — this issue is the decision.
1. Real outputSchema + structuredContent from the tool_schemas.py TypedDicts
Dropped originally for two reasons that are both gone: the FastMCP-1.x wrapper (replaced by SDK v2, which supports full JSON Schema 2020-12) and the token budget. The budget, however, has a hard constraint the re-evaluation must start from: **the advanced surface sits at 2
Read the thread · 2026-08-20 · closed · 3 comments
HVTrust evaluation for Openzim MCP: Grade B (70.8/100), ranked 23 of 718 MCP servers
Hi — I maintain HVTracker, an independent trust registry for AI agents and MCP servers. Projects are scored from public signals only (OSSF Scorecard, build provenance, signed commits, maintenance cadence, distribution reach). No self-reporting, no submission required, no paid placement.
Openzim MCP ranks near the top of the MCP servers we track, so I wanted to send you the result directly rather than just publish it.
[. The v2 port (#360, #362) deliberately kept --transport sse serving: the SDK still ships the transport, and removing a working transport is a user
Read the thread · 2026-08-13 · open · 1 comment
v2.5: tune zim_get descriptions for natural-language summary/structure/main-page dispatch
Background
The v2.0.0rc1 Stage E1 dispatch sweep (Qwen3-8B-Q4 against the 512-probe Gate 0b set, 5 reps) cleared the F2 enforcement gate (f2_pass=true, failures=[]) but three new-in-Phase-F operation classes showed low absolute dispatch accuracy:
| Class | Accuracy | n / 5 reps | Where the model goes instead |
|---|---|---|---|
zim_get-summary |
20% (20/100) | 20 probes | 80/100 → zim_query |
zim_get-structure |
53% (56/105) | 21 probes | 49/105 → zim_query |
Read the thread · 2026-05-27 · closed · 1 comment
[Feature]: Allow LLMs to query file list to avoid filling context
Pre-submission Checklist
- I have searched existing issues and discussions for similar requests
- I have read the project documentation and roadmap
- This feature aligns with the project's goals and scope
Problem Statement
If a filepath is not specified the project automatically spits out a list of available files. For a small list of <25 zim files perhaps this is fine, however when a query returns a list of 200, for instance, it wastes context space and user time. list_zi
Read the thread · 2026-05-01 · closed · external user · 1 comment
Most recent
zim_search: flat schema vs oneOf mode-conditional shape (Gate 0.3 re-run)
Split from #370, which closed part 1 (outputSchema/structuredContent) as not-planned on measured per-call cost. Part 2 is unrelated to that decision and is the only live content left, so it moves here.
The question
zim_search's input schema is flat: every mode's parameters are siblings, and which ones apply is described in prose. The alternative is a oneOf that makes the mode-conditional shape structural, so a client can only send a coherent combination.
The protocol constraint tha
Read the thread · 2026-08-25 · open · 0 comments
feat: Add binary content retrieval for embedded media (PDFs, videos, images)
Summary
Add the ability to retrieve raw binary content from ZIM entries, enabling integration with external tools for processing embedded media like PDFs, videos, and images.
Background
A user building a knowledge agent identified a gap in the current implementation: while openzim-mcp can discover embedded content (via extract_article_links which finds media links), there's no way to retrieve the raw binary data to pass to other tools for processing.
Many ZIM archives contain
Read the thread · 2026-01-28 · closed · 0 comments
The remaining reports are on the project's issue tracker.