# Give OpenClaw a Project and checked outputs. Reviewed: 2026-10-20 Canonical: https://clankercloud.ai/learn/use-openclaw-with-clanker-workspace Use Clanker Workspace as the Project and saved-output layer for an authorized OpenClaw task. Configure its documented native Streamable HTTP client, approve browser sign-in and prove access with a small task. Adding the connection does not host OpenClaw or start scheduled work in Clanker. ## What you are working toward One validated JSON result saved from a bounded processing task. ## Start with - An OpenClaw version documenting native MCP OAuth - A Clanker account and a fictional processing Project - Explicit approval for saved files and hosted compute ## Choose the transport explicitly The current documented configuration pattern is: openclaw mcp set clanker-workspace '{"url":"https://workspace.clankercloud.ai/mcp","transport":"streamable-http","auth":"oauth","oauth":{"scope":"workspace:read"}}'. Then run openclaw mcp login clanker-workspace and complete browser authorization. Keep the transport explicit: current OpenClaw docs describe SSE as the default when it is omitted. This example begins with Read; saving and computation require additional approved permissions. Sources: - [Clanker: connection, permissions and managed Sessions](https://clankercloud.ai/workspace-agents) - [OpenClaw: Streamable HTTP and OAuth](https://docs.openclaw.ai/cli/mcp/transports) ## Prove the connection before processing inputs Run openclaw mcp doctor clanker-workspace --probe, then ask the connected agent to call workspace_status. Confirm actual account permissions and discover the current native tool schemas. A configuration entry is not a connection test. Use the client’s documented reload or discovery behavior when previously cached tools no longer match. Sources: - [OpenClaw: connect MCP servers](https://docs.openclaw.ai/tools/mcp) - [Clanker: live agent instructions](https://workspace.clankercloud.ai/for-agents.md) ## Use a small input with known totals Place a fictional hours.csv in the selected Project: Engineering has 2 hours at 42 per hour; QA has 3 hours at 50. Ask OpenClaw to inspect the saved content, then prepare a reusable script producing processed/cost-summary.json. Expected totals are two rows, five hours and cost 234. Keep missing or malformed rows visible. ## Run one bounded job and inspect its receipt For this task, separately approve Edit for source/output saves and Configure and run for hosted compute. Update the client’s requested scopes through its documented OAuth configuration and authorize those permissions before proceeding; the initial read-only connection cannot execute the example. Start one managed Session, save the script and use the currently discovered saved-script tool with a timeout and declared JSON output checks. Follow that exact job to a terminal state. Do not relaunch it just because a response is delayed. Sources: - [Clanker: connection, permissions and managed Sessions](https://clankercloud.ai/workspace-agents) - [Clanker: live agent instructions](https://workspace.clankercloud.ai/for-agents.md) ## Validate the saved JSON rather than the promise Require the result to contain rowCount 2, totalHours 5 and totalCost 234. Check the job’s execution, validation and saving states separately. Read the actual saved JSON, retaining its file identity and revision, then open it in Workspace. Record the verified result in the Session before completing it. Failed or partial work needs an honest outcome. ## Keep the client, Workspace and compute separate OpenClaw stays wherever you run its gateway. MCP supplies approved Workspace operations; it does not move the gateway, channels or schedules into Clanker. A Project identifies task context, not a Project-only OAuth grant. BYOA hosting and standalone Sandbox API runtimes have separate setup, lifetime and persistence rules. This OpenClaw connection has not undergone authenticated Clanker acceptance here. Current official documentation establishes the setup path; actual ChatGPT and Claude tests do not prove this client/version. Finish the canary before expanding the workflow. Sources: - [Clanker: Projects, files and saved work](https://clankercloud.ai/workspace) - [Clanker: connection, permissions and managed Sessions](https://clankercloud.ai/workspace-agents) - [OpenClaw: Streamable HTTP and OAuth](https://docs.openclaw.ai/cli/mcp/transports) ## Try this prompt Paste this prompt into your connected OpenClaw session. The setup link opens the connection guide; it does not launch or send anything to the client. ```text Use your authorized Clanker Workspace MCP connection and current live tool schemas. First check status and permissions. In my fictional Project, inspect hours.csv: Engineering 2 hours at 42 and QA 3 at 50. After I approve Edit and execution, start one managed Session, save a reusable script and run one bounded job. Require processed/cost-summary.json with rowCount 2, totalHours 5 and totalCost 234. Inspect execution, validation and saving receipts, read back the saved JSON/revision, and complete the actual outcome. Do not publish, send messages, create schedules or request billing access. ``` ## Before using the result - The probe and authenticated status succeed on this client. - Only the intended saved input is used. - The job and saved JSON independently satisfy the declared totals. - Workspace shows the correct file and terminal Session result. ## Common questions ### Does this run OpenClaw inside Clanker? No. It connects an external OpenClaw client to Workspace tools. Running an agent harness on hosted compute requires a separate supported runtime setup. ### Can I reuse the ChatGPT connection grant? Create and approve this client’s own OAuth connection. OpenClaw’s default operator-managed credentials should not be mistaken for separate accounts for everyone using its gateway. ### Will this example run on a schedule? No. It is one bounded task. Recurrence requires a separately configured, authorized and tested scheduling arrangement. ## Sources - [Clanker: connection, permissions and managed Sessions](https://clankercloud.ai/workspace-agents) - [Clanker: live agent instructions](https://workspace.clankercloud.ai/for-agents.md) - [Clanker: Projects, files and saved work](https://clankercloud.ai/workspace) - [OpenClaw: connect MCP servers](https://docs.openclaw.ai/tools/mcp) - [OpenClaw: Streamable HTTP and OAuth](https://docs.openclaw.ai/cli/mcp/transports) ## Related - [/workspace-agents](https://clankercloud.ai/workspace-agents) - [An agent space with a place for the next step.](https://clankercloud.ai/learn/ai-agent-space) - [An agent runtime with a task you can inspect.](https://clankercloud.ai/learn/ai-agent-runtime) - [A cloud for agents, with the work kept together.](https://clankercloud.ai/learn/agent-cloud) - [Do the work. Build the tools. Keep it moving.](https://clankercloud.ai/workspace) - [Give Hermes Agent a Workspace for lasting work.](https://clankercloud.ai/learn/use-hermes-with-clanker-workspace) [Open Workspace](https://workspace.clankercloud.ai) [Current pricing and usage](https://clankercloud.ai/pricing)