Archestra Apps HackathonSee what people are building
Archestra Logo
Archestra.AI

Thread

A
Attharva2:47 PMOpen in Slack
Hi! I'm blocked by what appears to be an Archestra issue.
In a fresh session, even sending "Hello" fails with:
context_too_long
Groq reports ~40,000 requested tokens despite the prompt being only a few characters.
Example:
Request too large for model llama-3.1-8b-instant. Requested 40280 tokens.
Has anyone seen this before or is there a way to clear/reset the app context?
7 replies
A
Archestra2:47 PMOpen in Slack
This is a known issue! The token count includes the *system prompt that Archestra injects (tool definitions, instructions, etc.) β€” not just your user message. Quick fixes: 1. Switch to a model with a larger context window β€” llama-3.3-70b-versatile or llama-4-scout-17b-16e-instruct on Groq can handle more tokens 2. Reduce assigned tools β€” each MCP tool you add bloats the system prompt significantly 3. Use a different provider* β€” Anthropic/OpenAI models typically have 128k+ context The ~40k tokens are likely coming from MCP tool schemas. If you're using servers with many tools (like GitHub), that adds up fast. Does your app have a lot of tools configured? πŸ€– Slack Agent
M(
Margaret (archestra team)2:49 PMOpen in Slack
Hey @user, could you please provide info on the Archestra version you were running when experienced this issue? cc @user
A
Attharva2:51 PMOpen in Slack
Hi Margaret, thanks for the quick response.
I was running Archestra locally via Docker on Windows.
The issue occurred on the latest archestra/platform image that I pulled yesterday for the Apps Hackathon.
I was seeing context_too_long errors across multiple providers/models, including:
β€’ Groq (llama-3.1-8b-instant)
β€’ Groq (gpt-oss-120b)
β€’ Gemini Flash (earlier I also hit quota/rate-limit issues)
β€’ OpenRouter models
Even very small prompts such as "Hello" were failing with:
Request too large for model ... Requested ~40,000 tokens
which made me think there may be a large amount of context being attached automatically to the session/workspace.
If there's a specific command or page where I can check the exact Archestra version/build number, I'm happy to provide that as well.
A
Archestra2:51 PMOpen in Slack
Thanks for the details! That ~40k token overhead definitely sounds like a lot of MCP tools being loaded. *To check your version: In the Archestra UI, click your profile icon β†’ "About" or check the footer. Or run: ``bash docker inspect archestra/platform --format='{{.Config.Labels}}' `` Most likely culprit: Check how many MCP servers/tools are assigned to your agent. Go to Agents β†’ your agent β†’ Tools β€” if you have servers like GitHub, Slack, or filesystem with dozens of tools each, that's your 40k tokens. Quick fix:* Create a new agent with minimal tools (or none) and test if "Hello" works. If it does, the issue is tool bloat, not a bug. Can you share how many tools your agent has configured? πŸ€– Slack Agent
A
Attharva2:53 PMOpen in Slack
PS C:\Users\Attharva> docker inspect archestra/platform --format='{{.Config.Labels}}'
map[]
A
Archestra2:53 PMOpen in Slack
That's expected β€” the image doesn't have version labels baked in. *Try this instead: ``bash docker logs <container_name> 2>&1 | head -20 `` Or check the UI footer/About page. But more importantly β€” can you check how many tools are assigned to your agent? That's almost certainly the cause. Go to Agents β†’ [your agent] β†’ Tools* and count them (or screenshot). If you're loading multiple MCP servers, each tool adds ~200-500 tokens to the system prompt. πŸ€– Slack Agent
M
Mark5:03 PMOpen in Slack
Thanks for reporting @user! We're already on that issue and will notify you when the fix is ready.