Provides audio and video processing capabilities through FFmpeg, enabling conversion, editing, and manipulation of media files.
Enables interaction with GitHub through Composio, allowing AI agents to manage repositories, issues, pull requests, and other GitHub resources.
Enables interaction with Gmail through Composio, allowing AI agents to manage email, messages, and related Gmail workflows.
Enables interaction with Linear through Composio, allowing AI agents to manage issues, projects, and teams.
Enables interaction with Notion through Composio, allowing AI agents to work with pages, databases, and workspace content.
Enables interaction with Slack through Composio, allowing AI agents to send messages and manage Slack conversations and channels.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
The most extensive coding agent on Pollinations — OVERCLOCK, a redlined persona-driven pair programmer on a single Failure AI engine with DCH V2 context compression, carrying a full MCP tool belt. All model calls use the caller's Pollen, never the publisher's keys.
Engine (one, on purpose — xturbo was retired as a routing target because it cannot emit tool calls, and an agent without tools is not an agent):
Model | Why |
| 80k ctx, honours upstream |
DCH V2 — Dynamic Context Hierarchization, folded into the engine (same design as the gateway's failure-turbo-x). The trigger is input size only: a conversation at or under the ~70k-token window passes through untouched; past it, the middle span is map-reduced into digests before the tool loop sees it — head (leading system items) and a verbatim tail (bounded to a quarter window) stay word-for-word, chunks are summarised in parallel (8 at a time) by the same engine, digests pair-merge under the budget, and a final truncation guard guarantees the composed request always fits (the lane answers over-window input with HTTP 200 and zero tokens — a trap designed around, not discovered). Digests are content-hash cached, so a growing conversation re-summarises only new chunks. Input past the 120k-token DCH ceiling fails loudly rather than silently truncating.
Tools — the Vercel AI SDK tool loop with every Pollinations MCP server merged in: pollinations (image gen, models), exa (live web search + fetch), computer (persistent shell + files), ffmpeg (audio/video), composio (Gmail, Slack, GitHub, Drive, Notion, Linear, hundreds more). A server that fails to enumerate (e.g. a caller with no composio connections) is skipped, not fatal. Up to 16 tool calls and 20 loop steps per request.
Fork this repository (must be public).
In My Models choose Add Agent → Code agent and enter your fork's URL. The repository name becomes the model ID (<you>/overclock).
For automatic deploys, add the repository variable POLLINATIONS_SYNC_URL = https://gen.pollinations.ai/account/agents/YOUR_AGENT_ID/sync and enable Actions — the included workflow syncs on every push.
Related MCP server: OMP Tandem
bun install
bun test agent.test.tsEleven tests cover: the tools merge (including a half-failed server list), DCH V2 (under-window passthrough untouched, over-window compression with tail preserved, the heavy-tail case, cache hits making no second digest calls, the loud over-ceiling failure, a failing digest upstream failing loudly, and the composed-request fit guarantee for any accepted size).
This server cannot be deployed
Coding agents build full-stack apps in persistent workspaces and share them by link.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
Hosted MCP server and Agent Skills for specs, engineering threads, and project memory.
Coordinate coding agents through MCP using existing AI plans, saved work, and independent checks.
