Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
Airtable MCP Server
MCP server that exposes Airtable operations as tools for LLM agents. Uses StreamableHTTP transport on port 3000.
Source: domdomegg/airtable-mcp-server (cloned at build time).
Available MCP Tools
| Tool | Description |
|---|---|
airtable_list_bases |
List all accessible Airtable bases |
airtable_list_tables |
List tables in a base |
airtable_list_records |
List records in a table (with optional limit) |
airtable_get_record |
Get a single record by ID |
airtable_create_record |
Create a new record |
airtable_update_record |
Update an existing record |
airtable_delete_record |
Delete a record |
airtable_query_records |
Query records with filterByFormula |
Environment Variables
| Variable | Source | Description |
|---|---|---|
AIRTABLE_API_KEY |
K8s Secret mcp-airtable-credentials |
Airtable personal access token |
MCP_TRANSPORT |
Set by envMapping | Transport type (streamableHttp) |
PORT |
Set by envMapping | HTTP listen port (3000) |
AIRTABLE_MCP_LOG_TRAFFIC |
spec.env in mcpserver.yaml |
Enables inbound HTTP + outbound Airtable fetch logging in the pod |
Docker Build
docker build -t airtable-mcp-server:latest .
The Dockerfile clones the upstream repo, builds it, and runs as a non-root mcp user.
Kubernetes Deployment
Deployed via the McpServer CRD. The Context Mapper operator watches the CRD and creates the Deployment and Service automatically.
# 1. Create the secret (copy example.secret.yaml and fill in your API key)
kubectl apply -f secret.yaml
# 2. Apply the McpServer CRD instance
kubectl apply -f mcpserver.yaml
See mcpserver.yaml for the full CRD spec. Key fields:
- Image:
us-central1-docker.pkg.dev/your-gcp-project/clerum/airtable-mcp-server:latest - Namespace:
mcp-server - Transport: StreamableHTTP at
http://airtable-server.mcp-server.svc.cluster.local:3000/mcp - EgressBindings:
api.airtable.com:443/TCPso HCC can generate the required outbound NetworkPolicy - Resources: 128Mi/100m request, 256Mi/500m limit
Preferred network model: declare spec.egressBindings on the McpServer and let HCC generate the L3-egress NetworkPolicy. The local networkpolicy.yaml remains as a manual fallback template for non-HCC flows; render it through deploy/scripts/render-public-egress-exceptions.rb before applying so it receives the canonical public-egress CIDR exclusions.
When AIRTABLE_MCP_LOG_TRAFFIC=true, the image boots through a small local runtime shim that logs:
- inbound HTTP requests handled by the pod
- outbound
fetch()calls toapi.airtable.com - status/error timing without printing the Airtable token
Testing
Tests live in mcp-servers/airtable/__tests__/ and run from the parent mcp-servers/ directory:
cd mcp-servers
npm install
npm run test:airtable
Test suites (4 files):
airtable.config.test.ts-- CRD spec validation, secret references, transport configairtable.api.test.ts-- Mock Airtable client, CRUD operations, rate limitingairtable.mcp.test.ts-- MCP protocol operations (tools/list, tools/call)airtable.k8s.test.ts-- Expected Deployment/Service generation, resource mapping
All tests use mock clients and do not require Airtable credentials or a running cluster.