Accepting design partners

Secure MCP access for AI coding agents.

Bastion lets engineering teams connect MCP-compatible coding agents to GitHub, Jira, and Slack through a self-hosted MCP gateway with identity, permissions, policy enforcement, human approvals, rate limits, and audit logs.

Built for platform and security-conscious engineering teams adopting AI coding agents.

Request pilot access View architecture

[SELF-HOSTED] [GITHUB] [JIRA] [SLACK]

bastion gateway event stream

00:00:00.000 INFO [core] gateway ready 00:00:00.012 INFO [identity] agent=claude-code session=ses_4f2 00:00:00.045 INFO [connectors] github jira slack enabled 00:00:00.089 INFO [policy] loaded rules for repo actions ... 00:01:23.401 WARN [req_11a] high-risk action detected 00:01:23.405 INFO [approval] routing to human reviewer 00:01:23.512 INFO [audit] appended policy decision 00:02:45.118 INFO [approval] approved by reviewer 00:02:45.122 INFO [github] executing approved tool call

Problem

AI agents are getting tool access faster than teams can govern it.

Coding agents are moving from autocomplete to autonomous tool use. Once they can touch GitHub, Jira, Slack, or internal systems, teams need a central way to control what each agent can do, which actions require approval, and how every tool call is audited.

Module::Audit

Tool calls with an audit trail.

Every tool invocation is evaluated against defined policies before connector execution. Results are written to hash-chained audit logs designed for tamper-evident review.

audit-stream.jsonl

STREAMING

1>{"event":"tool.call","tool":"github.issue.list","decision":"allow"}
2>{"event":"policy.eval","tool":"jira.issue.create","decision":"allow"}
3|{"event":"approval.required","tool":"github.pr.create","decision":"suspend"}
4>{"event":"approval.ok","via":"slack","approver":"@sarah"}
5!{"event":"tool.deny","tool":"github.pr.merge","decision":"deny"}
6_waiting for next tool call...

Module::Core

Bastion sits between agents and internal tools.

Bastion sits between AI agents and internal tools, enforcing policy before tool calls reach company systems.

01 / AGENTS

MCP-Compatible Coding Agents

02 / GATEWAY

Bastion MCP Gateway

03 / CONTROLS

Identity + Policy Engine + Approval Workflow + Audit Log

04 / TOOLS

GitHub / Jira / Slack

Agent identity

Tool calls are tied to agent, client, and session context so teams can separate human access from agent access.

Policy enforcement before execution

Bastion evaluates allow, deny, approval, and rate-limit rules before routing a tool call to GitHub, Jira, or Slack.

Approval routing

High-risk actions can be paused for explicit human review before execution.

Hash-chained audit logs

Tool calls, policy decisions, approvals, and denials are recorded for review across the pilot deployment.

Secrets isolation

Connector credentials stay inside the gateway instead of being handed directly to agents.

Scoped permissions

Teams can shape agent permissions around the actions and tools they are ready to govern first.

Module::Connectors

Focused pilot connectors.

Bastion’s pilot scope focuses on the tools coding agents commonly need first: GitHub, Jira, and Slack. Policy, approval routing, rate limits, and audit logging sit at the gateway layer.

GitHub

Slack

Jira

Fit

Built for teams rolling out AI coding agents.

The pilot is for engineering organizations that need governance before expanding autonomous tool access.

01Your engineers use MCP-compatible coding agents such as Claude Code, Cursor, VS Code-based tools, or internal agents.
02You want agents to access GitHub, Jira, or Slack without giving them broad unchecked permissions.
03You need approval workflows for risky actions.
04You need audit logs for every agent tool call.
05You prefer a self-hosted gateway during early deployment.