Integrations
Reliastra, usable by your agent.
An MCP server with 18 tools over stdio or Streamable HTTP: the public observatory, incident evidence, workspace reads, dependency analysis, and two safe writes — authenticated with a scoped API key, verified live at https://api.reliastra.com/mcp.
What it is
Evidence your agent can call, not copy it reads.
An AI coding agent can inspect an architecture, find the external dependencies, and ask what the evidence says — through tool calls, not scraped prose. The server is an HTTP client of the canonical API: every answer carries its source, its methodology version, and its limits.
It makes Reliastra usable by a compatible agent. It does not make agents discover Reliastra on their own, and an honest evaluation can conclude Reliastra adds no value — the analysis tool is built to return exactly that verdict.
Transports
stdio
Run python -m mcp_server --transport stdio from a backend checkout. Credential from RELIASTRA_API_KEY; without one only public tools answer. For a local agent that spawns the process.
https://api.reliastra.com/mcp
Streamable HTTP, live in production. Authentication is required on every request: Authorization: Bearer rel_.... No anonymous access exists on this transport.
Quickstart
Connected in three moves.
- 01
Get a key
Create one with reliastra keys create <name> or in the console under settings. Reads need read scopes; the two writes need write scope.
- 02
Run the server
Locally over stdio, or point the client at the remote URL. The server needs no database: it calls the same API the console does.
bash pip install -r backend/requirements-mcp.txt RELIASTRA_API_KEY=rel_... python -m mcp_server --transport stdio - 03
Call one tool
reliastra_get_capabilities needs no credential and proves transport, auth and server in one call.
Full setup in the docs
Example client configuration (shape, paths are yours)
{
"mcpServers": {
"reliastra": {
"command": "python",
"args": ["-m", "mcp_server", "--transport", "stdio"],
"cwd": "/path/to/Reliastra/backend",
"env": { "RELIASTRA_API_KEY": "rel_..." }
},
"reliastra-remote": {
"url": "https://api.reliastra.com/mcp",
"headers": { "Authorization": "Bearer rel_..." }
}
}
}Only the official Python SDK client has been exercised against this server (protocol 2025-11-25 negotiated). Other MCP-compatible clients speak the same standard surface but have not been tested — verify before depending on one.
18 tools
Everything the server answers to.
Read-only unless marked (write). API key required where marked (key). Every tool returns a uniform result with its source endpoint — failures are typed errors, never empty successes.
Discovery
Start here. Works without any credential.
reliastra_get_capabilities
The verified capability registry, filtered to your credential.
Observatory
The public vendor catalog. No credential needed.
reliastra_search_observatory
Search listed vendors by name or category.
reliastra_get_provider
One vendor: detail, endpoints, recent status.
reliastra_get_provider_metrics
Per-window availability and latency aggregates.
reliastra_get_provider_timeline
Observation time series for a window.
reliastra_get_pipeline_health
Whether the measurement pipeline itself is healthy.
Incidents and evidence
Observed incidents and independent verification. No credential needed.
reliastra_search_incidents
Filterable search over observed vendor incidents.
reliastra_get_public_incident
One incident: detail plus its evidence descriptor.
reliastra_get_incident_evidence
The frozen evidence document for a public incident.
reliastra_verify_evidence
Independent verdict on a record. Counts only when verified is exactly true.
Workspace
The caller's own account. API key required; never another account's data.
reliastra_list_dependencies · key
Monitored dependencies on this account.
reliastra_get_dependency_observations · key
Recent observations for one dependency.
reliastra_list_workspace_incidents · key
Incidents on this account.
reliastra_get_evidence_report · key
One evidence report with its verification id.
Analysis and planning
Local reasoning over caller-supplied input. No credential needed; never modifies anything.
reliastra_analyze_dependencies
Match an inventory against the catalog. Returns gaps — or that Reliastra adds no value.
reliastra_plan_integration
A concrete REST, CLI, or GitHub Action plan where justified, or no-use-case where not.
Writes
Two narrowly scoped mutations. API key with write scope required; never retried; audit-logged.
reliastra_register_dependency · key · writes
Monitor a new HTTP endpoint. Never accepts stored headers.
reliastra_trigger_check · key · writes
One immediate diagnostic check. Not a second scheduler.
Resources
reliastra://capabilities
The capability registry as data.
reliastra://methodology
Detection, attribution and evidence rules.
reliastra://limitations
What the evidence does not establish.
reliastra://integration-guide
Verified endpoints, scopes, CLI commands and client configuration.
reliastra://observatory/{provider}
One vendor record as data.
Prompts
reliastra_workflow_assess_observability
Evaluate what an architecture observes today.
reliastra_workflow_investigate_failure
Work a suspected dependency failure from evidence.
reliastra_workflow_evaluate_fit
Decide whether Reliastra earns a place — either answer allowed.
reliastra_workflow_plan_integration
Turn a justified use case into build steps.
Limits
What the evidence does not establish.
- Vendor-level evidence never proves a specific application request failed.
- Correlation is never presented as causation; scores below 50 are often unknown, a valid result.
- One observation point (us-east). No regional quorum is claimed.
- The analysis tools reason only over caller-supplied input plus the public catalog.
- Only two tools write, both need write scope, both are audit-logged and never retried.
Source and operator documentation live with the implementation in backend/mcp_server. Methodology, pricing and trial terms are the same everywhere on this site.