Configure AI agents with organizational knowledge
Agent profiles are MCP prompts that give your AI tools a specific role, a set of standards to enforce, and clear instructions for how to respond — so every agent acts consistently and purposefully.
What are agent profiles?
Agent profiles turn a general-purpose AI into a focused team member.
An agent profile is a named system prompt stored in ContextRail and served through the MCP protocol. When your AI tool connects to ContextRail, every profile you have created appears as an available prompt. Selecting a profile loads its instructions into the AI session — shaping how the agent searches for standards, frames its responses, and decides what to flag.
Profiles pair naturally with contexts. A security reviewer profile tells the agent to retrieve security-tagged contexts before examining any code change. A docs writer profile pulls your writing standards and applies them to every documentation PR. Because the profile references live contexts, it stays up to date automatically as your standards evolve.
Profiles are organization-scoped. Admins create and maintain them; all members can use them. This keeps agent behavior consistent across the team without requiring each engineer to configure their own prompts.
Example agent profiles
These profiles illustrate common patterns — adapt them to your team's standards and naming conventions.
Enforces security standards during code review — OWASP rules, auth patterns, secrets hygiene.
Reviews UI code against design tokens, accessibility rules, and component patterns.
Checks API changes for REST conventions, versioning rules, and response shape contracts.
Walks new engineers through setup steps, tooling configs, and team conventions.
Evaluates structural decisions — service boundaries, data models, dependency direction.
Applies docs standards — tone, structure, terminology — when reviewing documentation PRs.
- You have access to a ContextRail organization.
- You have the admin role — only admins can create or edit profiles.
- Your AI tool is connected to your organization via MCP.
Creating an agent profile
Follow these steps to publish a profile and make it available to your team.
Decide what task or review scope this agent covers. A focused role — security reviewer, onboarding guide — performs better than a catch-all agent.
Describe the agent's goal, the standards it enforces, and how it should respond. Reference specific context IDs so the agent knows exactly where to look.
Link contexts the agent should retrieve by default. Use domain and tag filters so retrieval stays focused on the right standards.
Use a consistent prefix like reviewer- or guide- so agents are discoverable and easy to select in workflows.
Run the agent on a real PR or task. Check that it references the right contexts and produces actionable feedback. Refine the prompt until it consistently hits the mark.
- The profile appears in the Agent Profiles list with the correct name and description.
- Your AI tool lists it as an available prompt when connected via MCP.
- Selecting the profile retrieves the right contexts and produces focused, standard-aware responses.
- Create button missing: confirm you have the admin role in your organization.
- Profile not visible in AI tool: verify your MCP connection is pointing to the correct organization and token.
- Agent retrieving wrong contexts: check the profile prompt references the right domain, tags, or context IDs.