spranab

Thanks for the excellent references! We actually just added a SKILL.md following the agentskills.io spec based on your suggestion — it's live now: https://github.com/spranab/saga-mcp/blob/master/.skills/saga...

Great point about sqlite-utils FTS. Saga already has full-text search via tracker_search across all entities (projects, epics, tasks, notes), but we're using basic LIKE queries — proper SQLite FTS5 indexes would be a meaningful upgrade for larger projects. Adding it to the roadmap.

The bugwarrior/syncall direction is interesting too. A github_sync tool that pulls issues into Saga's local tracker (and optionally pushes status back) would close the loop between local agent planning and team-visible issue trackers. Definitely something worth exploring.

Re: SWE-bench agents and issue tracking — most SWE-bench agents read the issue description as input but don't write back to GitHub. Having a local structured tracker that an agent can read/write freely (without API rate limits or auth concerns) is exactly the gap Saga fills.

I built an MCP server that lets Claude (or any MCP client) orchestrate structured debates between multiple LLMs.

You give it a topic, it sends it to all configured models in parallel, they see each other's responses, refine their positions over multiple rounds, and a synthesizer produces a final consolidated output.

Works with OpenAI, DeepSeek, Groq, Ollama, Mistral, Together — anything with an OpenAI-compatible API.

npx brainstorm-mcp

Thanks for your valuable feedbacks.

SQLite handles this well in practice. saga-mcp uses WAL mode with busy_timeout=5000 and synchronous=NORMAL, so concurrent writes queue up rather than fail. The intended use case is one agent per project per session — if you had multiple agents writing to the same .tracker.db, WAL mode serializes the writes transparently.

For MCP adoption — it's growing fast. Claude Code, Claude Desktop, Cursor, and Windsurf all support it natively now. The spec is simple (JSON-RPC over stdio or SSE), so the barrier to both building and consuming MCP servers is low.

That actually is great, we can add ads detection and extract only the relevant information. Thanks @ksaj

What do you mean? I just wanted to share something I am working on. Trying to understand what you meant by ads.

Hi HN, I built SDF (Structured Data Format), an open protocol that sits between web content and AI agents.

The problem: Every agent that consumes a web page independently fetches HTML, strips boilerplate, extracts entities, and classifies content. A typical page is ~89KB of HTML (~73K tokens). When 100 agents consume the same URL, this extraction happens 100 times with inconsistent results.

What SDF does: Convert once into a schema-validated JSON document (~750 tokens) containing entities, claims, relationships, summaries, and type-specific structured data. Agents consume the pre-extracted representation directly.

Results from production deployment (2,335 documents, 10 content types):

99.2% token reduction from HTML 90% extraction accuracy with fine-tuned 1.5B + 3B model cascade 4.1x faster than monolithic 14B baseline Downstream experiment: general-purpose 7B model scores 0.739 accuracy from SDF vs 0.352 from raw markdown (p < 0.05) The pipeline runs locally on consumer hardware (dual RTX 3090 Ti). Fine-tuned models are open on HuggingFace (sdfprotocol/sdf-classify, sdfprotocol/sdf-extract). Protocol spec and JSON schemas are on GitHub.

Protocol spec + schemas: https://github.com/sdfprotocol/sdf Whitepaper: https://doi.org/10.5281/zenodo.18559223 Models: https://huggingface.co/sdfprotocol Happy to answer questions about the design decisions, the type system, or the evaluation methodology.