Skip to main content
Glama

AgentBridge Africa

Python Status License


What problem does this solve?

Agent systems interacting with real rails need hard cost ceilings, local context, and failure-mode safety. Soft retries become infinite. AgentBridge makes budget a hard stop and evaluates every change against golden trajectories before expansion.

Related MCP server: M-Pesa MCP Server

Architecture

flowchart LR
  A[ContextProfile<br/>locale / currency / rails / offline] --> B[Planner]
  B --> C[Worker<br/>MCP-shaped tools]
  C --> D[BudgetGuardian<br/>hard stop]
  D --> E[Verifier]
  E --> F[Golden evals<br/>including failure injection]

Stack: Python 3.11+ · Pydantic v2 · PyYAML · pytest
Eval harness runs fully offline.

What is implemented

  • ContextProfile (en-NG, en-KE, offline-NG)

  • BudgetGuardian hard stop

  • MCP-shaped envelopes with safety hints

  • Golden YAML trajectories + failure injection

  • Payments adapter (quote / execute / status)

Quick start

pip install -r requirements.txt
make eval

Demo

Demo GIF coming soon. Run make eval to see 100% on expected outcomes.

Portfolio value

Production-minded agent design for African contexts: hard budgets, regional profiles, deterministic evals.

Limitations

Planner is deterministic. Live multi-rail adapters are stubbed for eval safety.

Companion repositories

Visual assets checklist

  • Social preview (1280×640)

  • Demo GIF of eval harness

  • Logo in docs/assets/

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers