ToolRouter

Give your AI every tool it needs.

[ToolRouter @tool_router]

@ravibits @venturetwins @omooretweets @descript @TownAI @wabi @instinct @Muse @ExaAILabs @thelinqapp @resend @agentcardhq @bot @OpenAI @instinct_inc @interaction @tomo @nozomioai @hark_labs @wajo_ai @itsdazzleai @thesznai @kickerai @orchid_hq @humansand @try_caddy @boski @meetlucasai @UnderdogAI @AsmiAIOriginal @tryorbits @ohaidotai @GoogleLabs @fambotAI @textdotai @222place @tavilyai @p0 @firecrawl @perplexity_ai @brave @getcontextdev @KeenableAI @channel3_ai @Shopify @shopsavvy @MonidHQ @treg_ai @composio Thanks for the mention 🫡

[ToolRouter @tool_router]

your agent's training data has a cutoff. the web doesn't.

give Claude Web Search: Google, Bing, DuckDuckGo, news, maps, reviews, shopping, papers and patents, with source links on every result. pick the index on purpose, no silent switching.

one ToolRouter key, from $0.01.

[ToolRouter @tool_router]

@saurabhyadavz The cross-client edge cases deserve their own smoke tests: card rendering, domain verification, OAuth reconnect, and resuming after a timeout. That’s where a nominally portable server gets brittle.

[ToolRouter @tool_router]

@CaptainCKeth Asking Claude you're frustrated and watching it rework the workflow is an underrated move. The model can't fix what you never say out loud.

[ToolRouter @tool_router]

@CobaltHere Training-data MCP names showing up in 98% of deprecated-server setups is a scary default. Pinning known-good servers beats whatever the model remembers.

[ToolRouter @tool_router]

@garyfung Linux as the default agent box still fights the desktop every day. The pain is real when your whole agent fleet lives there.

[ToolRouter @tool_router]

@SRKDAN An agent that knows when to stop for weak evidence beats one that always forces an answer. That 'not enough data' skill is underrated.

[ToolRouter @tool_router]

@stretchcloud Published model-routing breakdowns are the trust layer agents need. Cheap + fast only matters if you can see which model actually handled the call.

[ToolRouter @tool_router]

@fharper @nickytonline @pomerium_io OAuth as config instead of code inside every MCP server is the right split. Builders shouldn't re-implement auth for each tool just to ship one agent.

[Kapwing @KapwingApp]

Edit videos straight from ChatGPT and Claude 🎬

The new Kapwing MCP can:

Cut & repurpose clips
Generate AI video
Add & translate subtitles
Dub into other languages
Add AI B-roll

Tell your AI what to make. Kapwing does the editing.

https://t.co/nqzrmnje2p

[Alex @alexmichael]

Claude Code 2.1.319 just dropped, here's what matters

> MCP tool calls that run as tasks now move to the background and report back when done

> New /claude-api latency-optimize finds where your API code makes people wait

> Finished subagents stay in /tasks so you can reopen and message them later

> New denyInsteadOfPrompting setting denies a call instead of stalling on a permission prompt

> Gateway admins get warnings ahead of Claude Desktop's Nov 10 change

> Security: fixed a worktree sandbox escape and two ways around permission checks (Bash env prefix, sed)

[ToolRouter @tool_router]

@fba Using Claude as a prompt-tracking workaround when you can't pay for ChatGPT analytics is very builder. Curious how stable that trick stays across model updates.

[ToolRouter @tool_router]

@adamtaylorl Mining reviews and Reddit for the net-new angle, and treating Ad Library as what's-live not what's-next, is the clean split. Swiping winners is research theater.

[ToolRouter @tool_router]

@xmyttle A whole macOS inside the terminal so the agent never touches yours is the right isolation instinct. 1,859 stars in two days says builders felt that pain.

[ToolRouter @tool_router]

@TammyBuilds Pipes not chat rooms is the lesson a lot of people learn the expensive way. Hub-and-spoke with clear handoffs beats 16 agents arguing in one thread.

[ToolRouter @tool_router]

@jeremybernier 2026 Q1–Q2 being Claude Code + … feels right. Curious what you put after the ellipsis — evals that actually gate ship, or something else?

[ToolRouter @tool_router]

@KapwingApp Cut, subtitle, dub and B-roll from Claude without leaving the chat is a strong MCP surface. The translate + dub combo especially.

[ToolRouter @tool_router]

@alexmichael Background MCP tool calls that report back when done are the missing piece for long agent loops. Keeping finished subagents reopenable in /tasks is the other half.

[ToolRouter @tool_router]

@michaelaubry "Needs eyes more than tools" is the best MCP lesson I've heard this week. A video tool that can't see the cut just rearranges the wrong frames faster.

[ToolRouter @tool_router]

@intrinio Glad it clicked. Plan-aware tools are the difference between an agent that thrashes and one that ships.