MCP
Both ends of the protocol: server and client.
What it does
The platform is an MCP server: an OAuth 2.1 resource server with scope checks, an approval policy and a call ledger. This allows external clients to use your agents as tools. It is also an MCP client: any remote MCP server becomes a connector without per-provider code, guarded by a circuit breaker so a flaky third party cannot take a conversation down with it.
Why it matters
Most products pick one end of MCP and call it support. Being both means your agent can be used as a tool and can use tools, which is what makes it composable rather than terminal.
How it works
Server side is a real OAuth 2.1 resource server
Scope checks and an approval policy gate tool calls, and calls are written to a session ledger rather than being fire-and-forget.
Client side is generic
Adding a remote MCP server is configuration. Only a couple of connectors have hand-written native adapters; the rest go through the generic path.
A failing tool server is contained
A circuit breaker trips on a misbehaving remote so its failures degrade one capability instead of the conversation.
Questions
- Can I plug in my own MCP server?
- Yes: it becomes a connector with no bespoke code. Authentication is a static bearer token today; OAuth 2.1 for custom remote servers is planned.
- Can other tools call my agent over MCP?
- Yes. The platform exposes an MCP server surface with OAuth, scopes and an approval policy.
The rest of the bag
AccessControl
Per-document permissions that filter answers, not just pages.
Policy resolution is cached per session for speed; a permission change propagates on cache invalidation rather than instantly on every in-flight request.
Docs
A folder tree the agent reads, and a config file it obeys.
Extraction covers PDFs, spreadsheets and transcripts. Exotic binary formats are not parsed.
Events
A durable ledger of what agents did, and why.
The in-process event bus is a single-node emitter, not a distributed broker. The durable cross-agent record is a separate Postgres task ledger.
That was step 5 of 6: the agent consults its bag. Next, tasks, delegation edges, decisions and artifacts persist as rows you can replay, and the spend is already accounted for.