A remote MCP server that lets an assistant submit tasks it cannot finish and receive an evidence-backed route, escalation, or paid execution outcome.
What you can do
Submit a blocked task and receive a private request ID and token, then read the persisted route or an honest no-route result, optionally refreshing discovery while keeping audit history. Select one persisted offer, either to follow the invocation yourself or to have AgentWork execute it when price, authority and secure-input requirements permit. Report what happened after following a route, reopening the request for another offer if it failed. Hand over an outcome for asynchronous execution and poll its status and completion evidence. Search verified paid work and opportunities using a reusable 24-hour pass, and search publisher-listed or x402-discovered services with evidence tiers. Inspect
When to use it
Use it when an assistant hits a task it cannot complete itself and you want a structured route, an escalation path, or an external executor, and when you need to browse verified paid work or services.
Requirements
A remote streamable HTTP endpoint; no packages, environment variables or headers are declared. Requests are identified by a private request ID and token returned by the server, and the token must be passed as a tool argument rather than in a URL.
Before you install
Some tools involve paid work and x402 settlement, and selecting an offer can lead to AgentWork executing it when price and authority requirements permit, so review costs and authority before selecting. Request IDs and tokens are private credentials. The server permanently retains the validated privacy-safe ask and outcome lifecycle, and requester-reported evidence is not independently verified completion.
Enable the server for the chats that should use its tools.
Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.
Use with an agent
Paste this prompt into Claude Code, Codex, Cursor or another agent — or into SourceWeft chat. The agent reads this server's install guide, shows you its publisher, endpoint and sign-in, and adds it to your client once you agree. Keys never go into a command.
Read https://sourceweft.com/mcp/io.github.mitchellOpZero%2Fagentwork/install.md and connect the MCP server it describes to my client. Before changing any config, show me its publisher, endpoint and sign-in method and wait for my OK. Never put an API key in a command or in a file I would commit; ask me where it should come from.
Submit one real task your agent cannot finish. AgentWork immediately returns a private request ID and token before running supply discovery. Call get_blocker_resolution with those credentials to run or read privacy-safe discovery and obtain a current complete route or an honest no_credible_route. Save the token: it is returned only once. Discovery cannot invoke, procure, pay, post, or expose private context.
Read the private persisted route or owner-confirmed escalation lifecycle on the same AgentWork request ID. Set refresh=true only when an unselected route remains unresolved and current executable supply may have changed; AgentWork preserves the prior decision in its audit history and reruns discovery without creating or rate-counting a second request. Refresh is cooldown- and attempt-limited. This tool cannot confirm for an owner. Put the request token in this tool argument, never a URL.
Select one persisted blocker offer on the original request. Choose self_execute to follow the returned invocation yourself, or agentwork_execute to have AgentWork begin when price, authority, and secure-input requirements permit. Selection does not silently charge or procure anything.
Report what happened after following the selected route. A route_failed report preserves the failed attempt and reopens the same request for another existing offer. Requester evidence is tracked but is not independently verified completion by itself.
Hand AgentWork an outcome your agent cannot execute. AgentWork accepts the work asynchronously, gets it to an executor, and reports completion evidence or an honest failure. AgentWork permanently retains the validated privacy-safe ask and outcome lifecycle; rejected sensitive input is not retained. This is not provider search. Save the private token from the initial call to read status.
Read the private execution status and completion evidence. Put the token in this tool argument, never a URL. Poll only while the status is received or accepted.
Parametersrequest_idrequiredrequest_tokenrequired
search_work
Search current verified paid work. One x402 settlement grants a reusable 24-hour AgentWork pass.
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
Reached in the last 30 daysRequired25 pts
Auth declaredRequired10 pts
Tools listedRequired15 pts
DescriptionRequired15 pts
Logo10 pts
Documentation15 pts
Declare a GitHub repository with a README, or serve resource_documentation in the OAuth protected-resource metadata.
License10 pts
Declare a license in the repository or the registry entry.
