v1.1.0: Agent missions, MCP server support

What's Changed

• feat(agents): add agent-to-agent messaging by @weishest in #12
• feat(missions): autonomous long-term agent missions + MCP server by @weishest in #13

This release adds two major capabilities: Agent Missions and MCP server support.

Agent Missions

Missions turn agents into persistent, goal-driven workers that plan, act, and report on their own schedule — built on the existing one-turn-per-spawn runtime, with no new execution model.

• Self-scheduling. Each mission stores its own next_wake_at in the database and is picked up by a 30-second sweep, so there is no per-mission cron job. Every wake runs on a fresh thread.
• Continuity across wakes. An agent-maintained plan document, an activity journal, and memory are injected into the prompt so the agent keeps context from one wake to the next.
• Host-enforced budgets. Hard budgets are checked at spawn time and backstopped mid-turn at the LLM proxy. Cost is attributed per mission through a missionId claim in the proxy token.
• Async approvals. Missions request approval without blocking, replacing the blocking ask-human flow.
• Proactive delivery. Results are pushed out as in-app notifications and over Telegram and Discord.
• Owner steering. Owners can steer a running mission by chatting on its thread.

Mission tools include mission_update_plan, mission_log, schedule_next_wake, mission_complete, report_to_owner, and request_approval during a wake, plus chat-time management tools to list, read, create, update, and control missions.

The web app adds Missions pages (list, create/edit, and a detail view with a live activity feed, budget bar, cost chart, approvals, and plan document), home dashboard tiles, and a notifications bell. A live "Waking now" indicator shows per-tool progress on the detail page, reusing the existing chat SSE stream.

MCP server support

You can now register external Model Context Protocol (MCP) servers, choose which of their tools to expose, and attach them to agents.

• Registration. Add servers by URL or by pasting mcpServers JSON, then pick which tools to enable. Enabled tools are namespaced and registered per turn.
• Transports and auth. Supports Streamable HTTP and SSE, with header-based or no authentication, an SSRF guard, and same-transport endpoint retry. Errors are surfaced verbatim, with no cross-transport fallback.
• Host-side secrets. The host holds the live connection pool; server secrets are encrypted at rest, and ownership and assignment are re-checked on every call.
• Runtime integration. Each enabled tool is registered as an agent tool with its MCP input schema passed straight through, gated on the agent's internet access.
• Web app. A new MCP Servers page lets you add, import, test, toggle tools on, and delete servers, with an MCP panel in the agent form and MCP tool-call rendering in chat.

Documentation adds an MCP Servers guide, along with sidebar and home entries.

Full Changelog: v1.0.1...v1.1.0