HasMCP
HasMCP turns an existing HTTP API or OpenAPI specification into a remote Model Context Protocol server without writing code, adding OAuth2 authentication, encrypted secret storage, token-pruning interceptors and real-time observability for LLM agents.
What is HasMCP?
HasMCP is infrastructure software that sits between an organisation's HTTP APIs and the LLM agents that want to call them. Its stated purpose is to close what the vendor calls the LLM-to-API gap, without the custom Python or TypeScript wrappers teams normally write and then maintain.
The entry point is a specification. You import an OpenAPI 3.0/3.1 or Swagger v2 file through the graphical interface, or define endpoints by hand, and HasMCP generates the tool schemas, parameter descriptions and typed payloads a model needs. The result is a remote MCP server exposed over Streamable HTTP, with no client-side installation and no configuration files.
The documentation organises the product around four objects. Environment variables hold credentials, either visible or masked, named in capital snake case and injected through a variable placeholder. Providers describe an external service through a base URL and its authentication, and receive an immutable secret prefix that isolates their variables. Provider endpoints become the individual tools a model can call, each with JSON schemas for its query arguments and request body. MCP servers bundle a provider and a selection of endpoints into a deployable unit with time-limited access tokens and a generated client configuration snippet.
Two capabilities carry most of the argument. Interceptors rewrite payloads before they reach the model: JMESPath expressions handle declarative filtering and projection, while an embedded Goja JavaScript engine handles stateful transformations and PII redaction with no Node.js cold start. Dynamic tool discovery, which the vendor calls the wrapper pattern, exposes only search tools and fetches full schemas on demand so that a large catalogue does not saturate the context window. HasMCP claims token reductions of up to 95 percent from this combination.
Authentication is handled natively through MCP OAuth2 elicitation: the server prompts the end user, exchanges the token just in time and injects it into upstream headers without exposing it to the model. An observability layer adds a streaming debug console, tool-call analytics, token economics and a payload inspector. Three products exist: a Community Edition under AGPL v3 that self-hosts on Docker with SQLite or Postgres, plus commercially licensed Pro and Enterprise builds on the managed cloud.
What it does
- Convert an OpenAPI 3.0/3.1 or Swagger v2 specification into typed MCP tools automatically
- Deploy a remote MCP server over Streamable HTTP, with nothing to install on the client side
- Prune JSON responses with declarative JMESPath expressions before they reach the model
- Run embedded JavaScript through the Goja engine to redact PII and reshape payloads inside the proxy layer
- Hold API keys, secrets and proxy headers in an encrypted vault the model never sees
- Authenticate end users against the upstream API through native MCP OAuth2 elicitation
- Watch tool calls live through a streaming debug console, call analytics and a payload inspector
When to use HasMCP / When not to
A quick filter to help you decide if HasMCP is the right fit.
When to use HasMCP
- Backend teams that already maintain an OpenAPI specification and want to expose it to an AI agent without writing a wrapper
- Platform and DevOps engineers who need to govern agent access to internal APIs through IAM groups, RBAC and audit logs
- LLMOps and AI platform engineers fighting context-window cost, who can lean on JMESPath pruning, Goja interceptors and on-demand tool discovery
- Security-minded teams that want API keys and OAuth2 tokens held in an encrypted vault and kept out of the model context
- Independent developers and small teams, who get a permanent 250-call monthly free tier and an AGPL v3 community edition they can self-host
When not to use HasMCP
- Anyone without an HTTP API or endpoint to expose, since the product has nothing to act on
- People looking for a chatbot or a language model: HasMCP is plumbing between APIs and agents, not an assistant
- Non-technical users, because the workflow runs on providers, endpoints, JSON schemas, environment variables and access tokens
- Organisations with formal compliance requirements, as no DPA, subprocessor list, SOC 2 or ISO 27001 certification and no GDPR statement is published
- Teams that must pin a hosting jurisdiction contractually, or that need a mobile app or an interface in a language other than English
How to use HasMCP
A typical end-to-end flow, from setup to results.
- Create an account on the hosted service, or install the community edition with Docker
- Open the Providers tab and create a provider with its name, base URL, description, REST type and visibility
- Note the immutable secret prefix the provider is given, since every variable it uses must start with it
- Store the credentials you need as environment variables, typed as visible or secret and named in capital snake case behind that prefix
- Add endpoints to the provider with their HTTP method, path and description, or bulk-import an OpenAPI v3 or Swagger v2 file
- Reference a variable inside an authentication header wherever the endpoint requires one
- Go to the MCP Servers tab, bundle the provider and toggle on the endpoints you want exposed as tools
- Generate a time-limited access token and copy the JSON configuration snippet from the server page
- Paste that snippet into your MCP client; free Claude accounts route through the mcp-remote command instead of a direct remote connection
- Add JMESPath filters or Goja interceptors, then follow the traffic in the debug console and the logs
Pros & Cons
Pros
- The shortest path on offer from an existing OpenAPI specification to a working remote MCP server
- Context optimisation is a headline capability rather than a setting, through JMESPath, Goja and on-demand tool discovery
- OAuth2 elicitation is handled natively, which is one of the most expensive parts of MCP work to build in-house
- Secrets stay out of the model context thanks to an encrypted vault, an immutable per-provider prefix and proxy-side injection
- Observability ships with the product, with live logs, analytics and a payload inspector instead of a third-party stack
- The AGPL v3 community edition self-hosts on Docker with SQLite or Postgres, which limits lock-in
- A permanent free tier of 250 tool calls per month allows a real evaluation before any commitment
Cons
- No GDPR mention anywhere on the site, no data processing agreement, no Article 27 representative and no subprocessor list
- No security certification is claimed, and no hosting jurisdiction or region is published for the cloud offering
- No postal address and no legal notice page: the company name appears only in a meta tag on the homepage
- A very young product, with the domain registered in May 2025, the first web archive in January 2026 and terms still at version 1.0.0
- Support runs through email alone, with no help centre or chat, and the enterprise route is an external Google form
- Headline performance numbers such as the 95 percent token reduction are vendor claims illustrated by screenshots, with no published methodology
- Strictly a developer product, since providers, endpoints, JSON schemas and access tokens cannot be avoided
Pricing & Plans
A permanent free plan is available. The Cloud Hobby tier includes 250 tool calls per month on a single user seat, with managed cloud hosting and pay-as-you-go overages, and the Community Edition is free under AGPL v3 for self-hosting. The lowest paid entry point is Cloud Pro at USD 24.00 per month, or USD 20.00 per month on annual billing, charged per seat and including 100,000 tool calls per seat. Enterprise pricing is quoted on request. No time-limited free trial is advertised.
- 250 tool calls per month
- one user seat
- managed cloud hosting and overages billed on usage
- 100
- 000 tool calls included
- high-availability clusters
- team collaboration and priority support
- billed per seat under the terms
- unlimited tool calls
- dedicated infrastructure
- custom SLAs and white-labelling
- arranged through a Talk to Founder form
- HasMCP Community Edition
- free and open source under AGPL v3
- self-hosted on the customer's own infrastructure
- HasMCP Pro and HasMCP Enterprise builds under a commercial licence
- with enquiries to license@hasmcp.com
Data, GDPR & hosting
A consolidated view of how HasMCP handles your data.
GDPR overview
There is no GDPR implementation to report. Across every page collected, from the homepage and the privacy policy to the terms of service, the contact page and the documentation, the words GDPR and data protection regulation never appear. No data processing agreement is offered, no Article 27 EU representative is named, no data protection officer is designated, no legal basis is stated and no subprocessor list is published. No SOC 2 or ISO 27001 certification is claimed either. Several practices point in a privacy-friendly direction without being framed as compliance: minimal collection, essential session cookies only with no third-party tracking or advertising cookies, IP addresses used for defensive thresholds and not linked to accounts, and a thirty-day purge after account deletion. Export and deletion requests go to privacy@hasmcp.com. The only regulator named anywhere is the American IRS, cited for billing retention.
Who owns the data?
The terms and the privacy policy leave customer data with the customer. HasMCP describes itself as a proxy: request and response headers and payloads are never stored, and user-defined interceptors transform them in memory. What the vendor does keep is account data, namely name, email and picture URL, with the email as unique identifier; provider Bearer and refresh tokens held encrypted in its database; and tool-call metadata used for billing and reliability, covering user identifiers, frequency and success, token optimisation and latency. IP addresses serve defensive thresholds only and are not associated with accounts. Account and API-key security is explicitly the user's responsibility, and the service is supplied as is.
Reuse rights
Data returned through the gateway belongs to the upstream APIs the customer connects, and the terms place no restriction on reusing it. The customer also writes the JMESPath and Goja rules that decide what leaves the proxy in the first place. The restrictions run the other way: users may not reverse-engineer the managed cloud infrastructure, evade tool-call limits by automating account creation, expose endpoints that break local or international law, or inject malicious scripts through interceptors. On the software itself, the Community Edition is licensed under AGPL v3 with the obligations that licence carries, while Pro and Enterprise are exempt from those restrictions for as long as a valid commercial licence is active. Neither document says anything about using customer data to train models.
Data retention & training
Hosting summary
No hosting jurisdiction is published. Neither the privacy policy nor the terms of service names a country, a region or a data centre for the managed cloud offering, which is described only as managed cloud hosting on Hobby, high-availability clusters on Pro and dedicated infrastructure on Enterprise. The single regulatory clue points to the United States, since aggregated billing data is retained for six years to comply with IRS rules. The site is fronted by Cloudflare, which tells you about the content delivery network rather than where data rests. No subprocessor list is published. What is documented is the handling rather than the location: HasMCP acts as a proxy, processes payloads in memory through user-defined interceptors, stores neither request nor response headers or payloads, streams debug logs to the browser without writing them to disk, and keeps provider Bearer and refresh tokens encrypted in its database. Teams that need control over location have an alternative in the AGPL v3 community edition, which runs on their own Docker infrastructure with SQLite or Postgres.
Things to keep in mind
Risks and trade-offs to weigh before adopting HasMCP.
- A single vault ends up holding the API keys and OAuth2 tokens of several services, so a compromised HasMCP account can expose everything upstream
- On self-hosted installs the documentation warns that losing your encryption key or secrets permanently blocks access to your own database, because nobody else holds them
- Exposing an internal API to an agent widens the attack surface, since a write endpoint becomes an action a model can trigger on its own
- Prompt injection travelling inside upstream payloads is illustrated by the vendor itself, and interception alone does not neutralise it
- An over-aggressive JMESPath filter silently removes data the model actually needed, with no visible error to warn you
- PII redaction is written by the customer in JavaScript, so one forgotten rule sends personal data straight to the LLM provider
- Usage-based overages mean a badly bounded agent loop can burn through tool calls without an obvious ceiling
Setup & Integrations
Technical difficulty
Moderate, and unavoidably technical. On the cloud edition the documented path from provider to endpoint to MCP server to client connection is announced at five minutes, with nothing to install client-side. Even so it assumes OpenAPI or Swagger literacy, HTTP methods and paths, authentication headers, JSON schemas and capital snake case variables. Connecting a client means generating a token and pasting a JSON snippet, and free Claude accounts must route through an extra command. Interceptors add JMESPath and JavaScript. Self-hosting is a step harder: Docker, a hand-built environment file, generated encryption keys, SQLite or Postgres, SSL and rate limiting.
Deployment
Integrations
Behind HasMCP
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement HasMCP.
Frequently asked questions
Do I need to write code to create an MCP server with HasMCP?
Which MCP clients can connect to a HasMCP server?
Is there a free plan, and what does it include?
How much does the first paid plan cost?
Can I run HasMCP on my own infrastructure?
How are API keys and secrets protected?
Does HasMCP store the data that passes through it?
How long is my data kept, and does the site address the GDPR?
Should you pick HasMCP?
HasMCP occupies a narrow, clearly drawn position: a piece of infrastructure between HTTP APIs and LLM agents, not a general-purpose product. Judged on that ground it makes a strong case. The route from an existing OpenAPI specification to a working remote MCP server really is short, and the parts teams usually underestimate, namely OAuth2 elicitation, secret handling and context-window cost, are treated as headline capabilities rather than afterthoughts. JMESPath pruning, Goja interceptors and on-demand tool discovery form a coherent answer to the token bill that large tool catalogues generate, and the built-in debug console removes the need for a separate observability stack. The AGPL v3 community edition matters as much as any feature, because a team that self-hosts on Docker keeps an exit.
The reservations concern the company rather than the software. Nothing on the site addresses the GDPR, no data processing agreement is offered, no subprocessor list or security certification exists, and no hosting jurisdiction is named for the managed service. There is no legal notice page and no postal address, and the company name surfaces only in a meta tag on the homepage. The product is also very young: the domain dates from May 2025, the first web archive from January 2026, and the terms remain at version 1.0.0. The performance figures, including the headline 95 percent token reduction, are vendor claims illustrated by screenshots rather than measured results anyone can reproduce.
The sensible approach follows from that split. Evaluate on the free tier or the community edition, where the technical claims cost nothing to test. Treat any deployment that carries compliance obligations as a separate question, and put the missing legal documentation to the vendor before it becomes a production dependency.
- Choosing a selection results in a full page refresh.
- Opens in a new window.