[HN Gopher] Show HN: Nia - MCP server that gives more docs and r...
___________________________________________________________________
Show HN: Nia - MCP server that gives more docs and repos to coding
agents
Hi HN, I'm Arlan, and I built Nia (https://www.trynia.ai), an open
MCP that integrates with coding agents like Cursor, Continue, and
Cline so they can retrieve external knowledge better than current
approaches. Coding agents generate code well but lose accuracy
when the answer lives outside the repo in front of them. Developers
end up pasting GitHub links, docs, and blog posts by hand and
hoping the agent scrolls far enough. Long context windows help, but
recent "context rot" measurements show quality still drops as
prompts grow. For example, in LongMemEval, all models scored much
higher on focused (short, relevant) prompts (~300 tokens) than on
full (irrelevant, 113k tokens) prompts, with performance gaps
persisting even in the latest models
(https://research.trychroma.com/context-rot). Nia is a MCP that
gives more context to any coding agent or IDE. It Indexes multiple
repos and docs sites and makes this available via MCP to your
coding agent so it has much more context to work with, giving you
more specific and accurate answers. Nia uses a hybrid code search
architecture that combines graph-based structural reasoning with
vector-based understanding. When a repo or documentation is
ingested, Tree-sitter parses it into ASTs across 50+ languages and
natural languages, and the code is chunked by function/class
boundaries into stable, content-addressable units. These chunks are
stored both in a graph db to model relationships like function
calls and class inheritance, and in a vector store. At query time,
a lightweight agent with give_weight tool dynamically assigns
weights between graph and vector search based on intent (e.g., "who
calls X" vs "how does auth work"), and both paths are searched in
parallel. Results are fused, enriched with full code context, and
passed through multi-stage rerankers: semantic reranker, cross-
encoders, LLM-based validators. Early Signal: In internal evals we
improved Cursor's performance by 27 % once Nia had indexed external
docs models couldn't get from their training data or searching the
web. Quickstart: <https://www.youtube.com/watch?v=5019k3Bi8Wo>
Demo: <https://www.youtube.com/watch?v=Y-cLJ4N-GDQ> To try it out:
grab an API key at https://app.trynia.ai/ and follow instructions
at https://docs.trynia.ai/integrations/nia-mcp. Try it and break
it! I'd love to know which contexts your agent still misses. Corner
cases, latency issues, scaling bugs. I'm here 24/7. Thanks!
Author : jellyotsiro
Score : 47 points
Date : 2025-07-24 15:05 UTC (7 hours ago)
(HTM) web link (www.trynia.ai)
(TXT) w3m dump (www.trynia.ai)
| NiloCK wrote:
| This looks interesting, and congrats on the launch. An immediate
| piece of critical feedback is that you should try to be a little
| more specific in the tagline. _All_ MCP servers give context to
| coding agents - that 's what an MCP server is (at least the
| resources/prompts channels of an MCP server).
| jellyotsiro wrote:
| thanks for the feedback! in this case, it is developer focused
| so primarily docs / external repos
| dang wrote:
| Ok, I've put that in the title above. Is is correct?
| jellyotsiro wrote:
| thanks!
| taherchhabra wrote:
| Sometimes Claude code defaults to using the older versions of
| certain libraries, have to explicitly tell claude to use the
| specific version. Even then it goes back to older version, so I
| downloaded the entire repo of that library and put it in my
| project folder. Does your product solve that ?
| jellyotsiro wrote:
| my product isn't for local projects and your own workspace but
| rather if you index other codebase, it will process it and make
| it callable using MCP (not on ur machine as files get deleted
| on runtime to prevent privacy issues)
| tevon wrote:
| Can you explain what you mean by "not on ur machine as files
| get deleted on runtime to prevent privacy issues". I may be
| misunderstanding, but I'd personally want the files to by on
| my machine, and served to my agent locally instead of being a
| remote MCP server that I don't have control of
| jellyotsiro wrote:
| Currently, the open source repositories and documentation
| you index are stored in both a graph database as embeddings
| (similar to Cursor.com). Indexing itself happens by
| creating a temporary file, which is deleted afterward.
|
| btw, I am working on allowing users to index their local
| files and fully store it locally! will update you on that
| kokanator wrote:
| I was unable to locate details regarding how the code/data is
| used/owned by the service. Clicking on the Legal link simply
| sends you to the top of the Home Page.
|
| At this time I can't even think about using the tool until I know
| what you are doing with my information and who owns or has access
| to it.
| jellyotsiro wrote:
| thank you so much, working on it
|
| edit: it is on the website now. forgot to add it, mb
| kissgyorgy wrote:
| [flagged]
| jellyotsiro wrote:
| hm, what model are you using? just ran it on my mac and got 100
| for performance and 96 for best practices.
| dang wrote:
| Please don't be a jerk on HN. We're trying for the opposite
| here.
|
| Especially please don't do this in Show HN threads, which have
| extra rules to forbid this kind of thing:
| https://news.ycombinator.com/showhn.html.
| afro88 wrote:
| > In internal evals we improved Cursor's performance by 27 % once
| Nia had indexed external docs models couldn't get from their
| training data or searching the web.
|
| What external docs do you have access to that aren't found on the
| web?
| jellyotsiro wrote:
| For evals, I used Skyvern -> https://github.com/Skyvern-
| AI/skyvern and their https://docs.skyvern.com/introduction
|
| LLMs and coding agents have general knowledge but they mostly
| give outdated info, even when asked to search on the web.
| nisegami wrote:
| How does this compare to Context7?
| jellyotsiro wrote:
| - nia can do deep research across any docs / codebase and then
| find any relevant links or repos to index.
|
| - it also supports both private and public repos :)
| efitz wrote:
| When I start a nontrivial coding task with AI, I added a
| "context" directory, instructions in the tool prompts how to use
| the files in that directory, and then I spent a couple hours
| using a thinking chat AI to generate the documentation I wanted
| (like "build me an API document for this library, the source code
| is at this URL and here are some URLs with good example code).
|
| I've had generally good results with this approach (I'm on
| project #3 using this method).
| jellyotsiro wrote:
| yep I used similar approach couple months ago but found it
| really inefficient because it took me some time
|
| give nia a try and use it on any docs, very curious to hear ur
| feedback
| luckystarr wrote:
| For thorny problems I let the agent give me a simplified flow-
| chart in mermaid syntax. LLM's brain-farts are easily visible
| then. I correct the flow-chart "Ah, you're right!" and then let
| it translate it to code. Works wonders.
| jellyotsiro wrote:
| that's smart! I've actually been thinking about integrating
| something like Mermaid flowcharts directly into Nia's output
| --visual context can make cursor etc understand context way
| better. have you found any particular types of problems where
| the flowchart approach really shines (or falls short)? Would
| love to hear more
| mrinterweb wrote:
| I often provide mermaid diagrams in my promt. Mermaid seems
| to be a good common markup to communicate relationships
| between humans and LLMs.
| stillpointlab wrote:
| I can confirm this is effective - I have done the same.
|
| I haven't done extensive experiments, but I have noticed
| anecdotal benefits to asking the LLM how they want things
| structured as well.
|
| For example, for complex multi-stage tasks I asked Claude Code
| how best to communicate the objective and it recommended a
| markdown file with the following sections: "High-level goal",
| "User stories", "Technical requirements", "Non-goals". I then
| created such a doc for a pretty complex task then asked Claude
| to review the doc and ask any clarifications. I then answer any
| questions (usually 5-7) and put them into a "Clarification"
| section. I have also added a "Completion checklist" section
| that I use to ensure that Claude follows all of the rules in my
| subdirectory "README.md" files (I have one for each major sub-
| section of code, like my service layer, my router layer, my
| database, etc). I usually go and do 2-3 rounds of Claude asking
| questions and me adding to the "Clarification" section and then
| Claude is satisfied and ready to implement.
|
| The bonus of this approach is I now have a growing list of the
| task specifications checked into a "tasks" directory showing
| the history of how the code base came to be.
| jellyotsiro wrote:
| oh damn interesting
| EcommerceFlow wrote:
| A killer app would be a up-to-date database of documentation from
| X amount of sources. For example, fully up to date Shopify API
| Documentation which could be included within cursor at the click
| of a button.
|
| I believe right now you're requiring us to do the
| scraping/adding?
| jellyotsiro wrote:
| great question:)
|
| Nia already supports that. Just take the link i.e
| https://mintlify.com/docs and ask to index it (it will crawl
| every subpage available from the rool link you specify)
| Fraaaank wrote:
| Context7 does this https://context7.com/
| jellyotsiro wrote:
| In my experience, it lacks a lot things that Nia can do:
|
| - nia can do deep research across any docs / codebase and
| then find any relevant links or repos to index. - it also
| supports both private and public repos :)
|
| lmk about ur experience with context7 (if u used) it and what
| docs did u use?
| hrpnk wrote:
| last time I checked, context7 depended on an opt-in from
| library authors expressed through a marker file in the
| repository, which is negatively affecting adoption and docs
| coverage.
| bdangubic wrote:
| or you can you know - use tools that are un-ethical but
| have better adoption :)
| smrtinsert wrote:
| Admittedly haven't tried documentation mcps at all, but can
| anyone quantify how much better it is than simply linking docs in
| the <LLM>.md file?
| jellyotsiro wrote:
| to just give quick comparison:
|
| one of my recent customers (yc s25) needed to migrate to stripe
| ASAP and cursor etc gave them deprecated docs. they used my
| tool to index entire stripe docs and then use it to migrate in
| couple hours:)
|
| lmk if u have more questions and happy to help
| fudged71 wrote:
| Is it 27% better or 10x better?
| jellyotsiro wrote:
| you can 10x your productivity haha :)
| gulercan wrote:
| we have a trademark for "Nia" in the UK. Keep in your mind when
| you need to scale.
| franky47 wrote:
| Having young kids, the Thomas the Tank Engine character came to
| mind.
| smcleod wrote:
| Do you plan on releasing the self hosted version free (or better
| yet - open source) for non-commercial, personal use?
| jellyotsiro wrote:
| Was thinking a lot about open sourcing MCP specifically.
|
| Will keep you in the loop
| smcleod wrote:
| My thinking is that across software development as an
| industry we have significant subscription fatigue. Obviously
| for hosted services there are ongoing costs involved - and
| regardless businesses should be paying people for their work,
| however individuals working on their own hobby projects and
| learning that often can't afford or prioritise the spend are
| the very people you want being advocates for your products
| within the commercial space. I think the middle ground for a
| lot of this subscription fatigue is to offer self hosted
| versions at no cost for non-commercial, personal use.
| jellyotsiro wrote:
| any companies that instantly come to mind that have similar
| structure you described? just curious (p sure browser use
| is one of them)
| whoodle wrote:
| I've been following you on X since your account was randomly
| shown to me. Your story is really cool, but also is your product.
| I'm working on my own LLM context problem as a side project (in a
| completely different space), but this could really fill voids I'm
| finding when doing it myself.
|
| I'm going to try this today. Best of luck with this!
|
| Are you still building this yourself (and Claude)?
| jellyotsiro wrote:
| that means a lot thanks! and your feedback would be golden, lmk
| how it goes.
|
| I suggest to watch this quickstart:
| https://youtu.be/5019k3Bi8Wo?si=3mMcp1Zd5C3Z0Rso
|
| Yes, I am building solo + claude code haha
| creepy wrote:
| What's the competitive advantage of Nia? What stops CursorAI or
| other companies implementing this feature themselves later?
___________________________________________________________________
(page generated 2025-07-24 23:01 UTC)