Both are MCP servers you add to Claude Code, and they are built for different jobs. Viktor MCP calls an agent that already exists and brings the answer back. Runbear MCP is where you assemble the agent itself.
Name check: viktor.com, the AI employee in Slack and Teams, is a different product from VIKTOR.AI (viktor.ai), an engineering platform with its own MCP beta. The tools here are viktor.com's.
Viktor MCP — what it's built for
Running an agent from outside your code, and collecting the result.
| Tool | What it does |
| ask_viktor | Open a thread, run it, wait for the result |
| create_thread · send_message | Start a thread and keep sending messages |
| wait_for_run · get_run · get_run_result | Check run state, receive output |
| cancel_run | Cancel a run in progress |
| list_threads · get_thread · list_messages · list_runs | Browse threads, messages, runs |
| get_file_download_url | Get a download URL for an output file |
| whoami | Check key scope and rate limit |
In short: an API for invoking an agent from another system and collecting the answer.
Runbear MCP — what it's built for
Assembling the agent: what it knows, what it can reach, when it runs, where it lives.
| Area | What you get |
| Bring your own skills | upload_local_skill_to_agent — push a skill you already wrote in Claude Code into the agent |
| Connect apps | search_apps, attach_app, detach_tool, Pipedream connections |
| Shape behavior | get_agent, update_agent, context blocks |
| Triggers and schedules | External event triggers, cron schedules |
| Search its knowledge | runbear_file_search across the knowledge base |
| Ask the user | runbear_ask_user_choice for button prompts |
| Ship to Slack | deploy_to_slack, join_slack_channels, file upload |
| Watch it run | list_recent_agent_traces, get_agent_trace |
In short: a management API for assembling, wiring, deploying, and watching an agent.
Viktor — run an agent and get the result. Runbear — build and configure the agent. Different halves of the same lifecycle, which is why they don't conflict.
Install both pieces
The MCP server manages agents. The plugin adds the commands that push your local skills into one.
claude mcp add --transport http runbear https://api.runbear.io/mcp
/plugin marketplace add runbear-io/skills
/plugin install skill-uploader@runbear-skillsNo key to paste — the first request opens a Runbear sign-in.
You decide what to build while you build it
This is the part a settings form is bad at. A form needs you to already know what you are making. What follows is one session in three frames — each frame is the same terminal a little further along, so the earlier lines stay on screen. The spec gets decided in the conversation, the skills you already wrote get attached one at a time, and the schedule and the Slack deploy are just more sentences.
Step 1 — the spec gets settled mid-session. The request lands incomplete on purpose: no rule yet for what counts as billing. A form would have to reject that. One question back, and the answer turns out to be a skill that already exists on disk.
Step 2, same session — the second skill goes up. Only the last two lines are new; the frame above is still there because nothing was restarted. Each skill is its own upload call with its own result, so one getting blocked would not take the other down.
Step 3, same session — the trigger and the deploy are sentences too. The weekday schedule and the #support channel were named here, not before the session opened. Read top to bottom, this one frame is the whole session.
Two skills that were sitting in a local folder are now part of an agent that runs every weekday and answers in Slack. Each upload is one call to upload_local_skill_to_agent; point it at a folder with several skills and it uploads them one at a time, reporting each result separately. The uploader is instructed to skip .env, secrets, and client config, and the server re-validates and can block an upload outright — a blocked skill comes back with a reason like secret_detected.
What actually transfers
Worth knowing before you point it at a folder you care about.
| What happens | |
| The skill file | SKILL.md lands under .claude/skills/ and loads on the agent's next activation |
| Several skills at once | One upload call each, reported per skill — a blocked one does not stop the others |
| Secrets | The uploader skips .env, .mcp.json, and client config; the server re-validates and can block with a reason |
| Anything a credential touches | Set it up in the dashboard vault, not the chat — that part is deliberately not conversational |
| A bigger folder | Beyond skills, /runbear:workspace-upload ships a whole directory into the agent's writable workspace |
And what stays in the dashboard, per Runbear's own limitations: model settings like temperature and reasoning effort, knowledge base sync setup, and scheduled jobs currently target Slack channels and DMs.
You never type a tool name
The same is true of the agent's own wiring. One sentence in, and the whole chain gets built without the user naming a single tool:
You wrote the first line. Every tool call under it is Claude Code picking Runbear's tools on its own — your client still asks before each write, and Granola sends you to a browser once to authorize. Integrations that need a stored secret are set up in the Runbear dashboard, not in the chat. Vendor docs, observed 2026-08-06.
And the result, five minutes after the meeting ended:
The agent you wrote in one sentence, five minutes after the call ends. Illustrative example.
Which one to install
Already running Viktor and you just need its answer inside a call — Viktor MCP. It opens the thread, waits, and hands the result back with no webhook and no trace lookup afterward.
Skills piling up in Claude Code that only you can run — Runbear MCP. That is the gap it closes.
Both — install both. They don't conflict, and they don't reach the same agent: Runbear MCP manages agents in your Runbear org, Viktor MCP delegates to Viktor.
Try it on something you already wrote
# once, if you have not already
claude mcp add --transport http runbear https://api.runbear.io/mcp
/plugin marketplace add runbear-io/skills
/plugin install skill-uploader@runbear-skills
# then, on a skill you already wrote
/skill-uploader:upload ./your-skill --agent https://app.runbear.io/agents/<appId>A Runbear account is free to start and needs no card. If you would rather begin from nothing, say this instead:
Create a Runbear agent called Support Triage that answers in a friendly, direct voice and escalates billing questions to #finance. Then show me what you made.
We build Runbear, so treat the framing accordingly. Viktor's table is its complete published tool set, condensed only in layout. Runbear's row for skill upload, knowledge search, and user prompts covers tools live in the MCP server that its public reference has not caught up with yet; everything else is in the published catalog. Observed 2026-08-06.
Sources, observed 2026-08-06: Runbear MCP · Runbear pre-built MCP servers · Pipedream Zoom trigger components · Viktor MCP · VIKTOR.AI, for disambiguation