MODEL CONTEXT PROTOCOL
Bring governed knowledge and decisions into MCP clients.
Governora exposes four Streamable HTTP MCP tools for governed retrieval and decision-aware work. The companion helps Claude, ChatGPT, Codex and other MCP clients, but the host still controls its own tool order, connectors and model calls.
The MCP companion provides search, fetch, prepare_task and record_task_outcome. It can supply permitted evidence and an execution contract, but it is advisory unless Governora controls the full managed lifecycle or receives a complete attested evidence packet.
What search covers
Search can return up to ten references from the active tenant boundary without calling a model.
- Workspace knowledge.
- Shared memory and automatic workspace context.
- Reusable artifacts.
- Indexed repository files.
What fetch returns
Fetch returns the decrypted content of exactly one reference ID that was previously returned by search. This two-step pattern keeps discovery separate from content access.
Prepare a task against decisions
prepare_task resolves exact active decisions for the task and returns compact MUST and SHOULD requirements before substantial work begins. It is not a generic knowledge search.
Record the task outcome
record_task_outcome attaches reported evidence to the prepared task and its exact decision versions. This preserves provenance without pretending the companion controlled every host action.
Workspace and organization scope
A workspace-scoped key is the smallest boundary and searches one workspace. An organization-scoped key can search workspaces in that organization. Gateway keys do not identify a person, so personal memory is never exposed through MCP.
Audit without raw query logging
Governora stores a tenant-scoped MCP audit event with the tool name, private HMAC fingerprint, result count, status and latency. The MCP audit record does not store the raw search query or fetched content.
Understand the advisory boundary
A SaaS AI host may call other connectors before Governora or omit Governora entirely. Its OAuth grants cannot be extracted or silently reused by Governora. Managed execution requires separate authorization or a customer-hosted runtime.
HOW IT WORKS
Four clear steps.
Each step stays connected to the same execution. This makes authority, external work and the final outcome explainable.
- 01
Prepare
Resolve applicable decisions and receive MUST and SHOULD requirements.
- 02
Search
Find and fetch permitted references without invoking a model.
- 03
Perform
Let the host execute work in its existing identity and tool context.
- 04
Record
Attach reported outcomes and evidence to the prepared task.
QUESTIONS
Simple questions. Direct answers.
What Governora does, what it does not do, and where it fits.
Can MCP access personal memory?
No. Gateway keys do not identify an individual Governora user, so the MCP search path excludes personal memory.
Does an MCP search call an LLM?
No. Search is a governed retrieval operation and does not send the query to a model.
Can ChatGPT connect directly today?
A remote endpoint must first be available over HTTPS. Public production ChatGPT publication also requires OAuth 2.1 or a private Secure MCP Tunnel; the current Bearer-key pilot is suitable for controlled client integration and local testing.
Can MCP change source systems?
No. Current MCP capabilities retrieve Governora knowledge and exchange decision evidence. Autonomous external write actions are outside the pilot scope.
Does companion mode provide the full Governora guarantee?
No. It is advisory because the host controls tool selection and execution order. Use managed gateway execution when Governora must guarantee planning before every external call.
Map one governed AI execution.
We will map the actor, decisions, capabilities, evidence, external calls and proof with you.
Plan a pilot review →