QualCoder MCP — analyse your QualCoder projects conversationally with Claude (open-source, now pip-installable, alpha)

[nicotem]

Hi everyone,

I've built an open-source tool that connects QualCoder to Claude via the Model Context Protocol (MCP), and I'd love feedback from this community.

Since you already use QualCoder, the short version: QualCoder MCP lets Claude work directly with your QualCoder projects — reading codes, categories, coded segments, cases, attributes and memos, and (with your approval at every step) coding new material, editing the codebook, writing memos and annotations, managing cases and attributes, and even proposing new codes from your data for you to review, refine and approve. It also exports your codebook, coded-segments reports and frequency/matrix CSVs, with counting rules checked against QualCoder's own source so the numbers match your Reports screen (and it says so when a number is computed differently). Works in Claude Desktop, Claude Code, or any MCP client.

It's now a one-command install: pip install qualcoder-mcp (v0.9.0-alpha).

How it relates to QualCoder's own AI: QualCoder already has excellent AI features, but they call commercial APIs you pay for per use. MCP is a different route — an open standard that lets an assistant like Claude work with your local tools and data — so you can use an existing Claude subscription to analyse conversationally instead of paying per API call. The researcher stays in control: the AI suggests, you review and approve, and only then is anything written; there are automatic backups, and writes are refused while QualCoder has the project open so the two can't clash.

It's free and open source (MIT), built against QualCoder 3.8.x and checked against QualCoder's own source for compatibility, and it's been through a whole-codebase security review. It's genuinely alpha, and the first software I've released — so please be gentle, and work on copies of your projects, never originals.

One important note: by design this tool sends your project content — including interview text — to Claude (Anthropic) for analysis. Please use synthetic or consented data and check your ethics/GDPR position before pointing it at real participant data — the repo's PRIVACY.md explains exactly what flows where.

I'd love testers and, just as much, methodological critique: I'm not advocating indiscriminate AI use in qualitative work; it won't suit every dataset or analytical style, but I think it's worth experimenting with. Bugs and feature ideas → GitHub Issues; a ⭐ helps others find it.

Repo: https://github.com/nicotem/qualcoder_mcp

(with thanks to the QualCoder team for the tool this builds on)

[qwertygc]

@kaixxx has developed significant updates for the AI ​​tab in QualCoder 4.0, including features similar to what you are working on. Perhaps you could reach out to him to join forces?

[nicotem]

Definitely, happy to help if I can, thank you for the suggestion!

[kaixxx]

Back from my summer break: This is interesting, indeed.
But as @qwertygc has already mentioned: It duplicates a lot of what has already been developed in QualCoder over the last couple of months. In fact, the current source code of QualCoder already contains an MCP server, it's just not opened to the outside world yet. But this was my intention from the very beginning, so that people could also use tools like Claude Code/Cowork together with QualCoder.

To get a first impression of the new AI agent inside QualCoder, look at my description here: #1383 Just like Claude, it also comes with skills (I call them "prompts" in QualCoder for legacy reasons), a sophisticated agent.md file, as well as a 3-stage permissions system and a lot of security guardrails that avoid having to babysit the agent and confirm every single step.

It would be great if you could help to test the QualCoder AI agent as it is now, as well as the MCP server once I open it to the outside world, which will happen soon. Currently, we are in feature freeze mode while preparing a new beta release, so I cannot do this immediately. But it's definitely on my list as a next step. If you can't wait, just create a fork and ask Claude to implement a simple network-connection to QualCoders internal MCP server via localhost. It should not be too complicated, as all the foundations are in place already.

Just looking at your readme, a lot of consideration seems to have gone into your development. It would be great if we could join efforts here, and I am very much looking forward to feedback and suggestions! I think we are very much on the same page: Welcoming a more "conversational" (or dialogical) analysis together with an AI agent, without giving up coding as a way to structure our insights systematically (see this preprint if you want to learn more: https://drive.switch.ch/index.php/s/HDCFkmmSd4xgWyP).