Building MCPBinder
I build and run MCPBinder through Reedstories. It brings project context, tasks, files, and decisions into a workspace that people and AI agentSoftware that uses AI to work toward a goal, choose steps, and use tools to carry them out. For example, an agent can look up project context and draft a task plan.Read more about AI agentAI agents (Software that uses AI to work toward a goal, choose steps, and use tools to carry them out. For example, an agent can look up project context and draft a task plan.) can use together. The design puts permissions and a readable record of changes alongside the work itself.
By Austin Howe · Updated
Select a dotted term for a short definition.
Keep the context with the work
An agent can act on an instruction, but the instruction rarely contains everything a project needs. Goals, constraints, earlier decisions, and the next useful action often live in separate conversations and files. Repeating that context takes time and makes it easier for the work to drift.
MCPBinder gives a project a shared record: its goal, context, constraints, decisions, and next actions sit beside the tasks and files. A person can update that record in the app, and a connected agent can read the parts its access permits before beginning a task.
Two ways into one workspace
People use the web app. Agents connect through MCP · Model Context ProtocolA standard way for an AI application to connect to tools and data. It defines how they communicate; each service still controls what the connection can access.Read more about MCP · Model Context ProtocolMCP (A standard way for an AI application to connect to tools and data. It defines how they communicate; each service still controls what the connection can access.), the Model Context Protocol. Both use the same project and task services, so an agent’s update belongs to the workspace the person already uses.
Tasks can move through configurable statuses and appear in a list or board. Project notes, files, and comments keep the supporting material close to the task. That makes the next step easier to find without turning every conversation into a new system of record.
Make access a choice people can understand
Connecting an agent is a separate decision from signing into the app. The person chooses the workspace, projects, and actions the connection may use. Access to a connected provider is also a choice; a workspace connection does not automatically grant access to every external service.
Connections can expire or be revoked. Broader permissions require fresh consent. I made these choices part of the experience because a useful agent connection needs a clear boundary around the work it can reach and change.
Make changes readable and attributable
A task status tells you where work stands. It does not explain why it changed. MCPBinder records activity and comments with the person or agent responsible, and gives people proposals they can review before accepting the proposed change.
Comments preserve the original attributed record. Corrections add to it instead of silently rewriting it. The aim is to leave enough context for someone returning to a project to understand what happened and continue the work.
A practical example: recurring research
Start with a project goal and the sources or files the research should use. Create a task that describes the question, the expected result, and the constraints. Then give the agent access to that project and the actions it needs.
For work that repeats, a recurring task uses a schedule and time zone to create the next task. The agent’s findings and attributed comment stay with the work. A person can review the result, record a decision, and update the next action before the cycle continues.
This illustrative workflow shows how the pieces fit together. The useful output is a project record someone can inspect and build on.
What the product brings together
MCPBinder combines project context, configurable task workflows, recurring work, files, decisions, and permissioned agent access. Those capabilities share a workspace and a history, so the handoff between a person and an agent can remain visible.
The lesson I carry into the rest of my work is straightforward: useful AI · Artificial intelligenceSoftware that performs tasks such as recognizing patterns, generating content, or making predictions. The projects here use it for research, learning, and creation.Read more about AI · Artificial intelligenceAI (Software that performs tasks such as recognizing patterns, generating content, or making predictions. The projects here use it for research, learning, and creation.) collaboration needs clear context, understandable permissions, and a result people can review.