Documentation Last updated March 14, 2026

AgentBridge Documentation

AgentBridge is a unified MCP (Model Context Protocol) gateway that lets AI agents talk to any integration — Notion, Slack, GitHub, Google Drive, and more — through a single secure API endpoint.


What is AgentBridge? #

Every MCP provider ships its own authentication flow, request format, and response shape. Connecting an AI agent to five tools means writing five different integrations, managing five sets of tokens, and handling five distinct error formats.

AgentBridge eliminates that complexity. Your agent calls one endpoint. AgentBridge handles OAuth, token refresh, request translation, error normalisation, and audit logging — for every provider — so you don't have to.

Your AI Agent  ──▶  AgentBridge Router  ──▶  Provider Adapter  ──▶  External API

Key Features #

Unified MCP Gateway
One POST /mcp/execute endpoint routes to any provider. Your agent doesn't need to know anything about the target API.
Secure Token Vault
All integration credentials are encrypted with AES-256-GCM at rest. Tokens are decrypted only at execution time and never exposed in logs.
Automatic Token Refresh
OAuth access tokens are refreshed transparently. Your agent never sees an expired-token error from a provider it has authorised.
Execution Logs
Every tool call is logged with latency, status, input params, and the full error body if the provider returned one.

How It Works #

  1. Authenticate with AgentBridge

    Generate an API key in the dashboard. Pass it as a Bearer token or configure it per-workspace.

  2. Connect a provider

    Users OAuth-connect a provider once (Notion, Slack, GitHub…). AgentBridge stores the encrypted token in its vault and refreshes it automatically.

  3. Execute any tool

    Call POST /api/mcp/execute with the provider slug, tool name, and parameters. AgentBridge retrieves the token, calls the provider, and returns a normalised response.

  4. Inspect execution logs

    Every call is logged. View latency, status, and full error detail in the dashboard or via GET /logs.

Execute a Tool #

POST /api/mcp/execute
Authorization: Bearer ab_live_xxxxxxxxxxxxxxxxxxxx
Content-Type:  application/json
json — request body
{
  "provider": "notion",
  "tool":     "create_page",
  "params":  {
    "parent_id": "abc123",
    "title":     "Meeting Notes"
  }
}
json — 200 response
{
  "status": "success",
  "data": {
    "id":  "page_456",
    "url": "https://notion.so/Meeting-Notes-page_456"
  },
  "latencyMs": 342
}

The user whose token is used is determined by your API key's owner. Each API key is scoped to a single account. For multi-user scenarios, create one API key per user or use the SDK's per-request user context.

TypeScript SDK #

bash
npm install agentbridge-sdk
typescript
import { AgentBridge } from 'agentbridge-sdk';

const bridge = new AgentBridge('ab_live_xxxxxxxxxxxxxxxxxxxx');

// Execute any tool through a single interface
const result = await bridge.execute({
  provider: 'github',
  tool:     'create_issue',
  params:  {
    repo:   'my-org/my-repo',
    title:  'Bug: login page broken',
    labels: ['bug', 'priority:high'],
  },
});

console.log(result.data);
// { id: 42, url: 'https://github.com/my-org/my-repo/issues/42' }

Supported Integrations #

Notion
Slack
GitHub
Google Drive
Gmail
HubSpot
Jira
Linear
Discord

View full provider docs, available tools, and OAuth setup guides on the Integrations page.

Authentication #

AgentBridge supports two auth methods for API calls:

Method Header When to use
API Key Authorization: Bearer <key> Server-to-server calls, AI agent integrations
JWT (session) Cookie: access_token=<jwt> Browser dashboard, first-party UI calls

Read the full Authentication guide to learn how to generate API keys, set scopes, and rotate credentials safely.

Plans & Limits #

Plan Price API Calls / mo Integrations API Keys
Free $0 100 2 1
Starter $19 / mo 5,000 5 5
Growth $49 / mo 50,000 Unlimited Unlimited
Enterprise Custom Unlimited Unlimited Unlimited

The Free plan never expires. No credit card required to get started.

Security #

  • Token Encryption — All integration tokens are encrypted with AES-256-GCM at rest and decrypted only at call time.
  • TLS everywhere — All data in transit is encrypted. HTTP is redirected to HTTPS.
  • Audit Logging — Every tool execution is logged with full request context, actor identity, and result.
  • Rate Limiting — Per-user and per-key limits protect against accidental flooding and abuse.
  • No log leakage — Parameter values are stored only in encrypted execution logs, never in plaintext application logs.

Read the Security overview for the full details.