Installation fails due to dependency conflicts
12 posts in the last 60 days, all from GitHub.
“Source Resume proof on the second Mac, 2026 10 12 (lead brief 2026 10 10 Part 1 step 5.4): fresh clone of main a99aef8 from a git bundle, python3.13 m venv .venv && pip install r requirements.txt , then the CI unit gate command from .github/workflows/tests.yml . Same 6 failure…”
“Summary Two independent problems in tools/sub agent.py (v0.6.3), both observed on a working OpenWebUI instance with remote MCP servers configured: 1. Every call to the tool fails instantly — TypeError: 'NoneType' object is not iterable when building filter identity . This…”
“The problem pyproject.toml declares mcp =1.0 with no upper bound. mcp 2.x removed mcp.server.fastmcp , which src/firm/mcp/tools.py:16 imports. Any environment built today gets mcp 2.2.0 and the test suite aborts during collection, so zero tests execute. Anyone running pip…”
Users report installation failures caused by dependency conflicts, particularly with Python packages and MCP modules. The issue stems from unbounded version constraints and breaking changes in upstream dependencies, leading to immediate failures on clean machines.
SOURCES (14)
“Source Resume proof on the second Mac, 2026 10 12 (lead brief 2026 10 10 Part 1 step 5.4): fresh clone of main a99aef8 from a git bundle, python3.13 m venv .venv && pip install r requirements.txt , then the CI unit gate command from…”
“Summary Two independent problems in tools/sub agent.py (v0.6.3), both observed on a working OpenWebUI instance with remote MCP servers configured: 1. Every call to the tool fails instantly — TypeError: 'NoneType' object is not iterable when building filter identity . This happens for any caller and any target model, so the tool is unusable as is on our instance. 2. MCP tools cannot be resolved for the sub agent — build tool server headers() is called with a keyword argument that no longer exists”
“context exploration engine/iowarp cei mcp/server.py : calls cte.MemContext() (lines 213, 780, 1276), which no longer exists in the Python bindings; requirements.txt pins only mcp =0.9.0 ; with current mcp ( =2) the server fails to start; unless CLIO SERVER CONF is set it generates a runtime config in an outdated format. The website's AI agents tutorial was pulled until this works. Found while running every example on the new website ( site/ ) against origin/dev 7bc16ccb4 in the devcontainer.”
“Summary Building the lza mcp server image from release/v1.1.0 currently fails at startup. The server crashes on import because the mcp dependency is unpinned and now resolves to mcp 2.x, which removed mcp.server.fastmcp (the module the v1 code imports). Any fresh build after mcp 2.0 was published hits this; builds done before that date would have resolved to 1.x and worked. Environment Branch/tag: release/v1.1.0 (tag v1.1.0 ) Build: make build (reproduced with both Finch and Docker) Host: macOS”
“My agent found this for me, after I encountered some issues running: npx y bci mcp Seems like this is fixed on main but somehow hasn't been published. Might just be a missed git push origin main tags step.”
“What bughunter/mcp/bughunter mcp/server.py and cli.py import from mcp.server.mcpserver import MCPServer , which only exists in mcp 2.x . But requirements.txt and the install docs pinned mcp =1.28 . Impact On mcp 1.28–1.30.x the bughunter mcp server fails at import ( ModuleNotFoundError: No module named 'mcp.server.mcpserver' ) → SystemExit(2) . mcp 1.28 is also the version flagged by the SkillsLLM scan (GHSA vj7q gjh5 988w, PYSEC 2026 3483). Repro Fix Bump floor to mcp =2.2.0 (matches the 2.x AP”
“mcp crosscheck version 0.1.0 Adapter mcpo Resolved client version 0.0.20 Target transport stdio Environment macOS 27.0, arm64, Node 26.5.0, Bun 1.4.2, uv 0.11.32 Minimal reproduction The README's first Quick start command, against the bundled fixture: Report or redacted output Expected behavior The default adapter set runs against a healthy server and reports its rendering. A client dependency break upstream does not fail every run. Actual behavior mcpo 0.0.20 declares mcp =1.17.0 with no upper”
“Summary infrastructure/mcp control plane/mcp config.local.json launches the memory sidecar with: This fails on a fresh clone: backend/pyproject.toml declares dependencies only in [tool.poetry.dependencies] (plus package mode = false ). uv reads the PEP 621 [project] table — which does not exist — so it resolves an empty environment , silently "succeeds", and then import mcp explodes with ModuleNotFoundError . Repro (uv 0.9.x, fresh clone of main @ 903a318) Side effect: uv sync also creates a str”
“The problem pyproject.toml declares mcp =1.0 with no upper bound. mcp 2.x removed mcp.server.fastmcp , which src/firm/mcp/tools.py:16 imports. Any environment built today gets mcp 2.2.0 and the test suite aborts during collection, so zero tests execute. Anyone running pip install cadre on a clean machine, including the maintainer on a new machine, gets a broken install. The existing development environments are green only because they froze mcp 1.28.1 months ago, which is what hid this. How it w”
“The break pyproject.toml pins mcp =1.0 with no upper bound. MCP 2.0.0 released 2026 07 28 and removed the Server.list tools() decorator, which src/shal/mcp/server.py:40 calls. Reproduced on a clean install (mcp 2.1.1): The published v0.2.1 wheel carries the identical pin , so this affects everyone who has installed since 2026 07 28 — the flagship capability (hand your rig to an LLM as gated tools) does not start. shal probe and shal tools still work; the break is contained to the MCP path becaus”
“Fresh install of berry 2.0.0 is broken: undeclared httpx dependency and incompatible mcp 2.x Summary A clean pip install "git+https://github.com/leochlon/pythea" (berry 2.0.0, 2026 08 28) produces a berry CLI that crashes on import, and — after fixing that — an MCP server that cannot start. Both are packaging issues in pyproject metadata. 1. import httpx is not declared Every berry invocation fails immediately: berry declares only mcp[cli] =1.0.0 and openai =1.0.0 . It used to receive httpx tran”
“Your current environment Environment OS: Linux Python: 3.12.13 vLLM: 0.27.1 MCP SDK: 2.0.0 (problematic) / 1.27.2 (working) Model: Qwen3 GPU: NVIDIA H20 96 GB NVIDIA Driver: 580.76.05 CUDA: 13.0 Installation: pip 🐛 Describe the bug Reproduction 1. Install the affected versions 2. Start any MCP SSE server A minimal MCP SSE server exposing one or more standard MCP tools is sufficient. The MCP server itself is not required to reproduce the API compatibility issue. The MCP server can be independent”
“Summary A fresh install per the README fails immediately. pyproject.toml declares mcp =1.2.0 with no upper bound, so pip resolves to the latest mcp (2.0.0). mcp 2.0.0 removed the mcp.server.fastmcp module that fusion cad mcp/server.py imports, so the installed console script cannot start. Because mcp 2.0.0 is now the default resolution on PyPI, this reproduces for anyone installing today. Reproduction pip reports Successfully installed ... fusion cad mcp 0.2.1 ... mcp 2.0.0 , then: Cause mcp.ser”
“Summary is contextvars broken() in sentry sdk/utils.py wraps the eventlet.patcher import in except ImportError: , but importing eventlet can raise AttributeError (not ImportError ) when trio is also installed. This crashes SDK initialization at import time. Environment sentry sdk : 2.66.1 Python : 3.14 eventlet : 0.41.0 greenlet : 3.5.1 trio : 0.33.0 httpcore : 1.0.9 dnspython : 2.8.0 OS : Arch Linux (rolling) Reproduction Minimal repro (no app code needed) Traceback Root cause The import chain”
