[Feature] MCP server for AI-native Multica orchestration from chat interfaces

[Korkyzer]

What

I built an MCP (Model Context Protocol) server that wraps the Multica CLI, enabling AI assistants in chat interfaces (Claude Desktop, Codex Desktop, etc.) to orchestrate Multica directly — create issues, assign agents, chain sequential workflows, monitor progress — without the user switching to a terminal.

Repository: https://github.com/Korkyzer/multica-mcp

Why

The CLI is excellent and covers everything. This MCP is complementary, for a specific workflow:

I find Claude Desktop (regular chat) is the best environment for brainstorming and architecture discussions, but I needed it to be able to open Multica issues, assign agents, follow progress, and chain sequential workflows directly from the conversation, without switching to a terminal. That's why I built this MCP.

What it includes

• 27 MCP tools covering issues, agents, projects, autopilots, runtime, attachments, workspace
• Companion SKILL.md with battle-tested best practices for AI agent orchestration (routing matrix, sequential chaining patterns, token optimization, anti-patterns)
• Token-optimized: compact JSON output, optional comment exclusion, pagination, configurable response windows
• Production-tested: used daily to orchestrate 10+ agents across multiple projects (470+ issues created in a single weekend)
• CI: lint + typecheck + unit tests (31) + build
• MIT licensed

Key features beyond CLI parity

• Fuzzy agent/project name matching with suggestions
• Agent/project name caching (5min TTL) to reduce CLI calls
• Safe issue creation patterns documented in SKILL.md
• Auto-convergent review loop patterns for upstream contributions

Integration options

Happy to contribute this however works best:

1. As a standalone package referenced from the docs
2. As packages/mcp in the monorepo
3. As an entry in an ecosystem/community tools page

Let me know your preference and I'll prepare the PR accordingly.

[benaltair]

Agreed this would be nice. I like the CLI-first philosophy though, similar to Pi. I just hope that any inclusion of an MCP server wouldn't nerf the work done on the CLI.

[emircanerkul]

I'm also looking into that. Instead Third party, well integrated MCP with permission configuration would be great instead saying agent to you only can assing agent a,b,c.

Also looking for native mcp server (not client). We could manage mcp servers within dashboard and deamon sync those and use it. It auto install via uv, it open up too much possibilities.

[yigithanyucedag]

Yeah, this would be nice 👍

[borgo]

@Korkyzer I am looking at ways to manage Multica configuration from an external agent, such as Claude desktop. Not so much to create tasks, more to manage the configuration of Multica itself: agents, skills, squads, etc. IMO, it is impossible to use the UI to change configuration manually at scale.

So far I have tried a "mirror repo" approach. I keep all the configuration as mostly pure MD files (with front matter) in a github repo. I created just scripts that are handle the upstream sync via the CLI (on remote server). Then the process of say adding new agent with 4 new skills is basically about changing files locally with claude code, creating a PR and once merged, asking the agent to deploy to my server (via CLI), and boom, Multica is changed.

I also created Multica admin agent, that can do all the config changes, using the same workflow via the repo. So all I have to do is ask Multica agent in a chat insude Multica UI, to change Multica's config. It handles all the above and it works surprisingly well too.

I'm not sure this is scalable across our teams though. Maybe an MCP in Claude desktop could make it more accessible for non-technical people to manage Multica's configuration for their workspace. But still, having a single source of truth in the repo, is quite nice.

I'm gonna try the MCP to get a feeling of how useful it is. Anyway, I'd appreciate your thoughts or advice about this. Maybe I am reinventing a wheel. Thanks.